rii track-1: data analytics that are robust and trusted

RII Track-1: Data Analytics that are Robust and Trusted (DART): From Smart Curation to Socially

Aware Decision Making

Strategic Plan

October 5, 2020

Award Number: 1946391 Jurisdiction: Arkansas

Start Date: July 1, 2020 End Date: June 30, 2025 (Estimated)

DART Strategic Plan 1

Contents 0. DART Strategic Plan ................................................................................................................................. 5

0.1. Introduction ........................................................................................................................................... 5

0.2. Mission ................................................................................................................................................... 6

0.3. Vision ...................................................................................................................................................... 6

0.4. Advancing the State of the Knowledge ............................................................................................. 7

0.5. Intellectual Merit ................................................................................................................................... 8

0.6. Broader Societal Impact ....................................................................................................................... 9

0.7. Overall Project Implementation ........................................................................................................ 10

0.7.1. Management ............................................................................................................................... 10

0.7.2. Roles and responsibilities within the Research Theme ......................................................... 12

0.7.3. Roles and responsibilities in the Project .................................................................................. 13

0.7.4. Sustainability .............................................................................................................................. 13

0.7.5. Post RII Track-1 Extramural Funding: .................................................................................... 14

0.7.6. Emerging Areas and Seed Funding ......................................................................................... 15

0.7.7. Partnerships and Collaborations: ............................................................................................. 15

0.8. Overall Risk Management Plan ........................................................................................................ 16

0.8.1. Challenges and Risks ................................................................................................................. 16

0.8.2. Mitigation Plans: ........................................................................................................................ 18

1. Coordinated Cyber Infrastructure ........................................................................................................ 19

1.1. Introduction ......................................................................................................................................... 19

1.2. Advancing the State of the Knowledge ........................................................................................... 19

1.3. Project Implementation ...................................................................................................................... 20

2. Data Life Cycle and Curation ................................................................................................................ 28

2.1. Introduction ......................................................................................................................................... 28

2.2. Advancing the State of the Knowledge: .......................................................................................... 28


3. Social Awareness ..................................................................................................................................... 37

3.1. Introduction (or Preface) .................................................................................................................... 37



4. Social Media and Networks ................................................................................................................... 55

4.1. Introduction ......................................................................................................................................... 55




5. Learning and Prediction ......................................................................................................................... 67



6. Education .................................................................................................................................................. 80



6.3. Broader Societal Impacts: ................................................................................................................... 80


7. Workforce Development and Broadening Participation ................................................................... 87



8. Communication and Dissemination ..................................................................................................... 95

8.1. Introduction ......................................................................................................................................... 95

8.2. Implementation Plan .......................................................................................................................... 95

9. Appendix A: Project SWOT Table ...................................................................................................... 101

10. Appendix B: Project and Theme-specific Logic Models .................................................................. 104


List of Acronyms ACDS Arkansas Center for Data Science ADHE Arkansas Department of Higher Education AEDC Arkansas Economic Development Commission AI Artificial Intelligence API Application Programming Interface APCD All-Payer Claims Database ARCC Arkansas Research Computing Collaborative ARGO The Great Plains Augmented Regional Gateway to the Open Science Grid ARP Arkansas Research Platform ASMSA Arkansas School for Mathematics, Sciences, and the Arts ASRI Arkansas Summer Research Institute ASU Arkansas State University BD2K Big Data to Knowledge BIGDATA NSF-NIH Interagency Initiative: Core Techniques and Technologies for Advancing Big

Data Science and Engineering CCPA California Consumer Protection Act CDS&E NSF program: Computational and Data-Enabled Science and Engineering CI Coordinated Cyberinfrastructure Research Theme; also Cyberinfrastructure CITI Collaborative Institutional Training Initiative COVID Corona Virus Disease CSTA Computer Science Teacher Association CUI Controlled, Unclassified Data DART Data Analytics that and Robust and Trusted: From Smart Curation to Socially Award

Decision Making DC Data Curation and Life Cycle Research Theme DG Data governance EAB Employee Advisory Board EAST EAST Initiative ED Education Research Theme EPSCoR NSF program: Established Program to Stimulate Competitive Research FIS a financial software company with offices in Arkansas GDPR General Data Protection Regulation GRA Graduate Research Assistant HBCU Historically Black Colleges and Universities HDFS Hadoop Distributed File System HIPAA Health Insurance Portability and Accountability Act of 1996 HPC High Performance Computing IHE Institute of Higher Education IUCRC NSF program: Industry-University Cooperative Research Centers LP Learning and Prediction Research Theme LSAMP Arkansas Louis Stokes Alliance for Minority Participation LSTM long short-term memory MTPP marked temporal point process NASA National Aeronautics and Space Administration NASEM National Academies of Sciences, Engineering, and Medicine NHPP Non-Homogeneous Poisson Process


NRT NSF program: NSF Research Traineeship Program ML Machine Learning MT Management Team OURRstore The Oklahoma University (OU) & Regional & Research Store PDC Positive Data Control PII Personal Identifying Information POC Proof of Concept RF Random Forest SA Social Awareness Research Theme SAC Science Advisory Committee; also known as the Arkansas EPSCoR Steering Committee SAU Southern Arkansas University SBIR Small Business Innovation Research SLURM Simple Linux Utility for Resource Management SM Social Media and Networks Research Theme SSC Science Steering Committee; also known as the Leadership Team STC NSF program: Science and Technology Centers STEM Science, Technology, Engineering, and Mathematics STTR Small Business Technology Transfer SURE Summer Undergraduate Research Experience SWOT Strengths, Weaknesses, Opportunities, and Threats UAF University of Arkansas, Fayetteville UALR University of Arkansas at Little Rock UAMS University of Arkansas for Medical Sciences UAPB University of Arkansas at Pine Bluff UCA University of Central Arkansas UGRA Undergraduate Research Assistant WCOB Walton College of Business, University of Arkansas, Fayetteville WD Workforce Development XSEDE NSF program: Extreme Science and Engineering Discovery Environment


0. DART Strategic Plan 0.1. Introduction

DART is part of a much larger State effort to establish an educational and research foundation for expanding and supporting a sustainable data science sector within our economy. Data science is targeted in this EPSCoR project because it is a strategically important technology for a significant and growing part of the State’s economy. Arkansas companies, including Walmart, Tyson, J.B. Hunt Transport Service Inc., Stephens Inc., First Orion, and Acxiom, make decisions from data, employ large numbers of data scientists, and are recruiting a workforce with a higher-level of broad and integrated data science skills.

Context is important here. The UAF Data Science B.S. program that serves as a basis for this project’s education component was developed because business leaders across the state - including those from the Fortune 500 companies named above - were pleading with universities in the state to help develop the data science savvy workforce necessary for them to remain competitive in fast-changing, data intensive markets. The pleas led to a coordinated plan among academia, industry, and government. At UAF, faculty from three colleges (business, engineering, and arts and sciences) developed - and the University invested in - an integrated Data Science B.S. program which has been approved by the Arkansas Department of Higher Education (ADHE) and began enrolling students in Fall 2020. The University also named data science as one of three Signature Research Areas and provided institutional resources (funding and support) for its development on campus. Industrial partners backed up their pleas with action and became active and vocal advisors of the data science program. State government contributed through the Arkansas Research Alliance (ARA), an organization founded by the Arkansas Economic Development Commission (AEDC) to attract research talent to the state. ARA helped fund the search and hiring of its newest ARA Scholar, SP Zhan, in the field of data science at UAF. In three short years, the coordinated efforts collected what was a dispersed set of data science courses and research projects scattered across colleges and departments into a coordinated program informed by a transdisciplinary team of faculty and facilitating convergent research on the campus. The trajectory of data science at UAF is a clear result of a coordinated effort between industry who expressed economy-driven needs, the academy which responded with a rigorous and substantial program to meet those needs, and state government that provided additional financial support and other institutional resources.

The necessary pieces are now in place across the state to replicate this trajectory: the 2018 Arkansas Science and Technology Plan identifies Data Science and Analytics as one of three Targeted Priorities in the state; data science programs are growing across the state - UCA offers a BS in Data Science (initiated by SP Addison), ASU, and UALR (SP Tudoreanu) all offer graduate certificates in Data Science, and similar but more focused programs exist at UAMS and UAPB; many senior faculty at participating institutions have robust data science research programs and junior faculty are looking for collaborators. DART, led by the state EPSCoR office and with a state-wide focus, completes the puzzle by providing necessary organizational structure and shared cyberinfrastructure. DART defines a coordinated research plan composed of five topical areas (Figure 1) related to data science needs identified by an expanded (beyond the UAF advisory board) number of industrial partners who drive a significant portion of our state’s economy. These topical areas are distinct but interrelated and are intentionally composed of researchers from different academic institutions across the state. Using the same relationships that made data science into a substantive, cohesive program at UAF, DART will establish a state-of-the-art data science research ecosystem involving multiple campuses across the state and develop a close interactive relationship between our university researchers and data science dependent industries. This relationship will expand the number of high-paying data science jobs in Arkansas and the resulting expansion of the data science sector of our economy will depend on a large continuous flow of well-trained data science capable talent. DART will include support for programs degrees and curriculum extending from middle school through graduate school to support this workforce pipeline.


Figure 1: DART research themes and economic drivers. Research to remove barriers to better data driven science and decision separated into five themes and informing four areas of economic activity.

0.2. Mission

To improve research capability and competitiveness in Arkansas by creating an integrated statewide consortium of researchers and educators working to establish a synergistic, statewide focus on excellence in data analytics research and training.

As Arkansas transitions to a more diverse, data-driven economy we must create an environment for university and industry collaborations in data science that will sustain this new economy with cutting-edge research and educate a workforce that enhances competitiveness in Arkansas industries. By bringing together experts from different data science sub-fields and application areas, we expect to develop both specific and comprehensive solutions that would be difficult to obtain in isolation. Collaboration with our industry partners provides a better definition of both problems and solutions in data analytics and workforce education.

0.3. Vision

The Arkansas research community - academic, government, and industry - collaborate often and easily on a shared computing platform with access to high performance computing nodes, peta-byte scale storage, fast and reliable big data transfer, and shared software environments which facilitates replicable, reproducible, and cutting-edge data science research. Reliable, scalable, explainable, and theoretically grounded data science approaches to data life cycles and modeling allow the public to better understand how machine learning and artificial intelligence effects their lives. When they engage with data science products on their smart devices, on social media platforms, and on the web, the improved and robust privacy and safety protections and fair results increase their trust of data collection and the resulting information, allowing for broader use of data science to benefit society. In Arkansas, the educational ecosystem provides learners with a well-designed, consistent, sequenced, scaffolded, and modular education in data science with further educational and/or job opportunities available at appropriate points in their careers.


0.4. Advancing the State of the Knowledge

The growing array of tools - powerful high-level programming languages, distributed data storage and computation, visualization tools, statistical modeling, and machine learning - along with a staggering array of big data sources, has the potential to empower people to make better and more timely decisions in science, business, and society. However, there remain fundamental barriers to practical application and acceptance of data analytics in these areas, any one of which could derail or impede its full development and contributions.

1. Big data management: Before data streams and datasets can be used in the many kinds of learning models, they are often manually curated, or at the least, curated for a specific problem. We still rely on hosts of analysts to assess the content and quality of source data, engineer features, define and transform data models, annotate training data, and track data processes and movement.

2. Security and privacy: Government agencies and private entities collect and integrate large amounts of data, process it in real-time, and deliver products or services based on these data to consumers and constituents. There are increasing worries that both the acquisition and subsequent application of big data analytics are not secure or well-managed. This can create a risk of privacy breaches, enable discrimination, and negatively impact diversity in our society.

3. Model interpretability: Machine learning models often sacrifice interpretability for predictive power and are difficult to generalize beyond their training and test data. But interpretability and generalizability of trained models is critical in many decision-making systems and/or processes, especially when learning from multi-modal and heterogeneous big data sources. There is a continuing to need to better balance the predictive power of complex machine learning models with the strengths of statistical models to better configure deep learning models to allow humans to see the reasoning behind the predictions.

These three barriers form the integrative research questions on which DART will focus. Activities in each research theme contribute to the integrative questions and the degree of interaction between themes is defined by that joint contribution. The flow of research from the research activities defined below, through the integrative questions, and finally into economic development through industry partners is captured in the Figure 2. Flow capacities are based on the number of objectives/activities from each theme that contribute to the integrative question and the, still admittedly arbitrary, importance of each barrier to economic activity in state.

Figure 2: Sankey diagram showing 1) the contributions of each research theme to the fundamental barriers we are addressing as a project and 2) industry dependence on research to removing or mitigating the barriers. Research theme collaborations and integrations are designed to address these barriers from different research approaches and perspectives.


0.5. Intellectual Merit

The Learning and Prediction research theme supports this through the creation of novel statistical learning methods in big data environments that are equipped with capabilities for addressing heterogeneity and hidden sub-populations within big datasets. Specifically, we will create statistical learning methods in big data environments that are equipped with capabilities for addressing heterogeneity and hidden sub-populations within big datasets. In addition, contributions will be made in mode specification and interpretation through the contribution of efficient variable selection using non-parametric methods. Lastly, we will advance computing in big data environments for traditional statistical modeling through statistical computing performed on distributed/parallelized computing nodes. Holistically, our theme will address challenges surrounding high-dimensional, dynamic and unstructured data sets and explore solutions in the domains of genomics, transaction scenarios in eCommerce, and supply chain logistics.

Data Life Cycle and Curation goal of building a “machine” that can analyze and manipulate data as well as a person (a data analyst) is challenging. A data analyst brings a tremendous amount of experience and knowledge into the process, and representing, storing, and expressing this level of knowledge and experience will stretch the current capabilities of AI technology. While a general data washing machine robot that will work for any dataset might be decades away, creating useful and scalable solutions for these three particular uses cases (data cleaning, data integration, and data tracking) is an achievable goal within the 5-year time frame of the grant.

Social Awareness research theme will greatly advance socially aware data analytics and sharing by 1) researching and documenting privacy breaches, security concerns, and discrimination in big data applications and understanding factors leading to those negative outcomes; 2) producing a suite of novel technologies, differential privacy preserving, attack resilient, secure multi-party computation, and crypto based mechanisms/algorithms for a variety of data acquisition and analysis tasks; 3) conducting cut-edge research in socially ware crowdsourcing, user-centric data sharing in cyberspace, cross-media discrimination prevention via multi-modal deep learning, fairness aware marketing strategy design, and privacy-preserving analytics in health and genomics; and 4) creating a Web portal that includes policies, regulations, practices, algorithms, tools, prototype systems, and a collection of publicly available datasets and real data from our business partners.

Social Media and Networks primarily includes 1) innovative methods, techniques, and platform for mining argumentation data and analyzing its characteristics, such as polarization, opinion diversity, participant influence, opinion community, and opinion prediction; 2) creation of a transformative multilayered network analytic method of analyzing deviant behaviors in social media networks by modeling multi-source, supra-dyadic relations, and shared affiliations among deviant groups; 3) multimodal deep learning methods to work with multimedia data from social media and other data platforms; and 4) innovative algorithms for logistics planning in disaster response using big social data analytics.

The Arkansas Research Platform (ARP) will push the edge of distributed high-performance computing coupled to distributed high performance storage via high bandwidth networks. While all these commodities are generally available at larger institutions, they are often out of reach for smaller institutions. Smaller institutions that do manage to acquire small compute clusters outgrow them quickly. The goal of the ARP is to federate these scattered resources into one whole resource. As important as the lessons learned from the ARP experience is the improved access to cyber infrastructure to researchers in the state of Arkansas, which will facilitate research, particularly big data analytics, that would have been out of reach to many researchers scattered across the smaller institutions within the state.


0.6. Broader Societal Impact

DART, as a center, is integrating data science research across the State and creating a deep and diverse data-ready workforce, which will pay immediate dividends in the form of increased federal grant funding and increased industrial research funding. As the State better aligns its investments with industry strengths, more opportunities to improve the quality of life in Arkansas and steadily increase educational attainment and wages will develop. Each thematic research area contributes in complementary ways to this mission.

Big data analytics is a heavy consumer of compute and storage resources. The lack of access to such resources acts as a barrier to talented researchers from under served and smaller institutions. ARP intends to flatten that playing field by giving all researchers at Arkansas institutions regardless of size or budget access to the compute and storage resources available at the larger institutions. Past experience has shown that having such access can greatly increase the pace of discovery by tapping intellectual resources that otherwise would be under-utilized due to a lack of access to adequate compute and storage resources. While access to the resources through ARP is crucial, it will not have a significant impact if its clients lack the technical skills to make effective use of them. Expanded data science undergraduate and graduate degree programs are necessary but smaller, more focused training on how to build research code and tools using the platform are equally important and will translate to industry and government environments. Organizations like The Carpentries offer well developed and tested training modules on basic modern computing tools (Git, IDEs, markdown), high-level programming libraries, visualization tools, and data science libraries necessary for effective data science.

Currently data scientists are spending only 20% of their time modelling, and 80% of their time working on cleaning and preparing data. Reducing that 80% would significantly increase productivity. Even bringing this down to only half that (40% effort to clean / prepare data) would be of enormous benefit to both industry and research in Arkansas, allowing data scientists to spend more of their time (80% vs. 20%) on working with data analysis and modelling.

Research contributions from improve the learning and prediction of data in a spectrum of applications including commerce, cybersecurity, disaster and emergency management, energy, environment, healthcare, retail, and transportation. Research outputs will generate interest in data science and help engage, encourage, and recruit a broad spectrum of learners as well as researchers. As a result, Arkansas should see a growth in research and education initiatives in data science. We expect to grow the segment of Society that can benefit from Artificial Intelligence-driven solutions by eliminating economic barriers to technology access and boost Artificial Intelligence applications and efficient platforms to support Arkansas economy and workforce development.

This research will address security and privacy, to practical application and acceptance of data analytics, and develop novel, integrated solutions for achieving privacy preservation, fairness, safety, and robustness in big data learning and sharing. The proposed research will help organizations and individuals to be aware of the uses, benefits, and risks of big data, determine whether disclosure of private information, unfair treatment, or potential risks have occurred or would occur, and assist community in the endeavor to provide trustworthy technologies. The principles, methods, tools, datasets, and evaluation results will have significant effect on development of socially responsible science and engineering workforce in Arkansas. Moreover, by advancing socially aware data analytics and proposing viable solutions that will assure that big data are collected and used in a safe, private, fair and responsible way, this project will contribute to the wider acceptance and support for big data products. Finally, innovative methodologies and tools developed for socially aware learning and sharing will help U.S. companies compete and lead globally.

Include understanding the increasing societal polarization, amplified by the massive reach of “always on” social platforms, that is threatening and damaging democracy around the world. The models and


insights generated will enhance our ability to both capitalize on the potential of social media as a force of good and mitigate its use as a weapon. Threats to democracy are abated through new models to understand how polarization forms, methods to detect online deviant behaviors, and interventions to prevent the spread of misinformation and rise of echo chambers. One of the direct applications of the research is in disaster management. Extreme weather events and major natural disasters are ranked by world leaders as the biggest risks facing our planet. The research will benefit disaster response decision-making by affording new tools and technologies that extract, classify, index, and analyze diverse and semantically rich multimedia social data to boost situational awareness. SM theme engages diverse faculty and students to develop smart, explainable, and accurate data analysis techniques.

Integrating data science research across the State and creating a deep and diverse data-ready workforce will pay immediate dividends in the form of increased federal grant funding, increased industrial research funding, and increased employment of well-paying jobs. As the State better aligns its investments with industry strengths and needs, more opportunities to improve the quality of life in Arkansas and steadily increase educational attainment and wages will develop. The HDR Big Idea recognizes that efforts in developing data cyberinfrastructure, education programs, and a deep workforce are most effective when linked to relevant data science research.

For the first time in AR EPSCoR history, we are working with three of the state’s Historically Black Colleges and Universities (HBCUs) to ensure inclusive learning pathways for a broad population of students. The HBCU roles are critical to the success of the entire education component.

DART will include a data science summer institute for undergraduates, summer internships and research experiences, increased data science educational opportunities, and revamped curriculum to include relevant data science topics and capstone projects.

0.7. Overall Project Implementation

0.7.1. Management A senior management team oversee research activities within and across the topic areas. It will be the

responsibility of this team to ensure that collaborative activities are ongoing, productive, and fall within the defined research goals of the project.

ARP will be managed as a unique multi-institutional resource. The CoLeads will serve as the Leadership Team for this resource and direct the management of resources at their respective sites. Dr. Cothren will serve as the executive director. The leadership team will define the operational procedures for the ARP combined resource in consultation with a user committee comprised of major users from each campus. A memorandum of understanding among the campuses participating in ARP will define the governance structure and establish operational parameters. This governance and operations model is based on our experience operating the established facilities at UAF, UAMS, and UALR. Direct support to research faculty and students will be provided by existing staff at UAF and UAMS, with additional faculty and students from UALR assisting in the development of testbed solutions.

Our management plan reflects the previous EPSCoR project organizational structure. It includes a state-supported Central Office which provides general oversight for the project and coordinates interactions with state boards and agencies; a management team (MT) comprised of administrators from participating campuses, Table 1, to ensure project implementation on campuses and information flow; a researcher-led Science Steering Committee (SSC) provides oversight for the scientific aspects of the program; and one or more external advisory boards contribute stakeholder perspective and facilitated dissemination of results to other groups.


Table 1: Management team comprised of administrators from campuses receiving subawards.

Name Role Institution Steve Stanley PI, EPSCoR PD Arkansas Department of Commerce Jackson Cothren Co-PI, Science Director University of Arkansas, Fayetteville Brian Berry Administrator, MT University of Arkansas, Little Rock John English Administrator, MT University of Arkansas, Fayetteville Tom Risch Administrator, MT Arkansas State University Angela Barlow Administrator, MT University of Central Arkansas Shuk-Mei Ho Administrator, MT University of Arkansas for Medical Sciences Mansour Mortazavi Administrator, MT University of Arkansas, Pine Bluff Anthony Johnson Administrator, MT Philander Smith College George Herts Administrator, MT Shorter College Abdel Bachri Administrator, MT Southern Arkansas University

Central Office: Dr. Steve Stanley, Arkansas NSF Program Director (PD) will serve as the project PI/PD. He is responsible for the overall statewide project management, which includes administration of the central office, overall management of science, outreach, and workforce development efforts, cyberinfrastructure, and evaluation, developing new funding opportunities to leverage or add support and linkage with other federal grants, and providing progress report to and communicating with NSF EPSCoR Office. Cathy Ma, Assistant Director and Program Administrator (PA), is responsible for day-to-day program operations, assisting with financial transactions, organizing events, and managing project staff. Jennifer Fowler, Director of Education, Outreach, and Diversity will expand the education and workforce development capacity of the project and coordinate activities to broadening participation. Fowler also administers the reporting database (ER Core) and communication activities. By leveraging Arkansas Department of Commerce staff as management of Arkansas NSF EPSCoR projects, we ensure a statewide focus with direct access to state government leadership, state agencies and private sectors, the Governor’s office, Accelerate Arkansas leadership, and other major statewide research efforts including the ARA. Under this arrangement, we can better coordinate travel and statewide meetings, leverage statewide resources to provide expanded training and/or workforce development efforts, and expand public engagement.

Succession Plan: Tom Chilton, Division Director of Science & Technology at Department of Commerce, and Chase Rainwater, Associate Professor of Industrial Engineering at the UAF, will take the responsibility of the PI and Co-PI, respectively, in the event that Stanley and Cothren cannot perform their duty.


Figure 3: Management structure of DART.

Figure 4: Summary of faculty and affiliation by research theme.

0.7.2. Roles and responsibilities within the Research Theme Each faculty and staff participant in DART will be assigned one or more roles in DART. Individual

assignments are based on research expertise and interest and designed to foster collaboration and increase communications across the research themes, cyberinfrastructure development, and education and outreach components. Information about roles and responsibilities will be posted on the project website throughout the project.

Research Theme CoLead: A CoLead has two important administrative roles: 1) serve as liaison between SSC and the respective research theme faculty groups; 2) lead the respective research theme in maintaining synchronous progress across all activities within the research theme and integrated research themes, including but not limited to: coordinating meetings and distributing communications among research theme faculty members; communicating progress to the SSC; communicating upcoming or existing personnel vacancies to the SSC. The CoLead role is primarily an administrative role and does not include any additional resources or support.

Research Theme Activity PI and CoPIs: Each research them includes several activities. For each activity there is one PI. The PI should communicate with the activity team regularly to ensure the project is performing satisfactorily. The PI should liaison with the research theme CoLead. In cases where


multiple CoPIs lead a research theme activity, it is their collective responsibility to perform the duties of research theme activity PI.

Research Theme Sr. Personnel: In each research theme activity, a number of faculty within that theme are assigned as senior personnel. They each have particular expertise and skills needed to accomplish the milestones in that activity. Faculty may be senior personnel in multiple activities and many also serve as PI or CoPI in other activities.

Research Theme Liaison: Liaisons have the task of disseminating information between research themes. Individuals with this role have subject area expertise that is relevant in multiple themes and across multiple activities. While they are not necessarily responsible for milestones in their cross-theme activity, they should: 1) attend meetings and keep up with progress in that activity, and 2) inform their theme of relevant results and possible collaborations.

0.7.3. Roles and responsibilities in the Project Science Steering Committee (SSC; also known as: Leadership Team): Comprised of CoLeads from

each research theme. Provides oversight for the scientific aspects of the program and is responsible for ensuring research theme milestones and objectives are being met annually. The SSC is also responsible for participating in NSF Site Visits, annual conferences, and communicating progress to the external evaluation board and external evaluator via annual reports and presentations. The SSC works closely with the EAB, PI, and CoPI to provide technical and/or scientific guidance as lead researchers on the project. Each SSC member is responsible for planned research in the theme and planning, execution, reporting, and dissemination via inter-institutional workshops.

Management Team: Comprised of vice-provost level administrators from each campus receiving a subaward. The PI, CoPI, and Management Team are responsible for financial decisions and other administrative duties.

Science Advisory Committee (SAC; also known as the Arkansas EPSCoR Steering Committee): Committee is composed of representatives from academia, government, and the private sector. The SAC selects the topical areas for each Track-1 Project, designates the fiscal agent/proposing organization as the responsible recipient for the RII Track-1 award, and must provide support for the Track-1 Project for NSF acceptance.

External Advisory Board (EAB): The EAB will include researchers from peer and aspirant universities or national labs who serve as technical consultants providing recommendations on research progress and strategic and long-term sustainability planning during annual site visits. The EAB will serve a critical role in the seed grant program as well as in mentoring and commercialization efforts.

Industry Advisory Board (IAB): The IAB members will serve as an intermediary between academia and industry. The IAB will include representatives from Arkansas industry sectors who will be impacted by DART research. Ex-officio members include the project Co-PI (to communicate scientific results), the project PI (to serve as the liaison to government and policy organizations), and members from related organizations like the Arkansas Center for Data Science (ACDS) and the ARA. The IAB will meet annually in conjunction with the Annual All-Hands Meeting, and they will meet quarterly to review results and recommend new areas of research and collaboration based on industry needs. One member of the IAB (rotating annually) will serve as a member of the EAB during site visits.

0.7.4. Sustainability DART is an important part of Arkansas’ plan to transition to a knowledge-based economy, and the

gains made through the NSF EPSCoR Track-1 projects are too precious to be lost for lack of support. Arkansas has concentrated its priorities for state investments and has built-in methods to provide resources to encourage and maintain the gains from prior investments. DART will utilize a variety of


sustainability strategies including leveraging of other existing programs, partnerships with private industry, user fees for facilities, and transitioning of operations to other agencies.

0.7.4.a Sustainability of Project Activities: The Arkansas S&T plan targets research areas where the state has both an academic research strength

and an industrial and workforce need. Because the activities and infrastructure in this project are intentionally aligned with the S&T plan, DART will receive funding from future state investments from other state agencies, educational institutions, and organizations as described in the S&T plan. DART is expected to produce many valuable products and services with the potential to improve research capabilities in Arkansas, including improved data science infrastructure (hardware, software, cyber, human), methods, algorithms, scientific knowledge, partnerships, and intellectual property.

Arkansas has created a large number of state incentives and investment programs to support targeted research and the commercialization of its products. As part of the state’s commitment to supporting data science in Arkansas the Governor, legislature, and our industrial partners established the ACDS. With funding from the State and its private industry members, ACDS has a long-term mission of promoting and supporting data science research, education and job creation in Arkansas. ACDS and a number of state incentives targeting technology-based research and commercialization will be available to support DART activities that have proven to be effective. We anticipate that DART will become the research and development arm of ACDS, and plans are in motion to secure annual research funding for this goal.

0.7.4.b Physical and Cyberinfrastructure Sustainability As part of its long-term support for data science education, research, and industries, a goal of ACDS

is to make statewide access to a shared infrastructure that includes HPC, storage, and data science tools more readily available. The infrastructure planned in DART will help connect not only our large research universities but many of our smaller universities and colleges that have very limited research activities. DART infrastructure and the ARP will be given priority for support by ACDS and the State in their future investments.

0.7.4.c Human Infrastructure Sustainability DART recognizes the importance of maintaining human infrastructure investments including

retention of faculty and students, continuing their career development, and maintaining industry relationships. A number of strategies proposed by DART will be sustained directly by ACDS, State, and Federal funding. ACDS plans to continue support for DART student and faculty research fellowships, internships, and stipends for faculty training to expand the number of course offerings available in data science in Arkansas, as described in section 4.4. We also have plans to seek funding to support the Arkansas Summer Research Institute (ASRI) through training and education grants at NSF.

DART will hold regional and virtual workshops to enhance the research competitiveness of the state’s faculty in data science and computing. Workshops topics will include science, grantsmanship, and commercialization. These will be tailored for students and early career faculty and will feature program officers from various agencies, entrepreneurs, and outside scientists. The state also provides small grants to allow young technology-based start-up companies to pay for the cost associated with applying for SBIR funding.

0.7.5. Post RII Track-1 Extramural Funding: Federal funding priorities are coalescing around data science. A number of Federal programs will be

targeted by DART researchers for sustaining extramural activities. A few examples include the following NSF programs: Industry-University Cooperative Research Centers (IUCRC), the Science and Technology Centers (STC): Integrative Partnerships, the Computational and Data-Enabled Science and Engineering (CDS&E) initiatives, and the NSF Research Traineeship Program (NRT). NIH recently released a strategic


plan for data science, with the goal to advance NIH data science across the extramural and intramural research communities. DART will investigate NIH programs such as Big Data to Knowledge, as well as interagency programs like the NSF-NIH Interagency Initiative: Core Techniques and Technologies for Advancing Big Data Science and Engineering (BIGDATA), as well as data-focused programs at other agencies like U.S. Department of Energy, U.S. DoD, and NASA.

0.7.6. Emerging Areas and Seed Funding Over the course of the award period, we anticipate project researchers and advisors will identify

emerging or transformative areas of research worthy of support or previously unidentified opportunities. To respond to these opportunities, the Management Team, in consultation with the External Advisory Board and the Industry Advisory Board, will allocate up to $318K in year 1, $380K in years 2, 3, and 4, and $300K in year 5 in seed grants to support emerging areas of research. A request for proposals will be posted online and distributed to all Arkansas institutions, with emphasis on collaboration. Proposals will be reviewed by a panel made of up of the EAB and IAB members. The duration of each project can be from 12 months to 24 months. Areas chosen for support will match the project’s scientific focus and support one or more of the project goals. That is, they will be emerging opportunities that strengthen ties to Arkansas business; enhance our talent pool by expanding and/or leveraging research collaborations; or provide unique educational or training opportunities that could strengthen the currently planned programs and lead to new cross-disciplinary efforts not previously proposed. We will target exploratory research projects and unique, newly emerged areas of transdisciplinary education, diversity, or industrial outreach. Included in these funds for researchers and students for allocations on commercial clouds (Azure and Google). The management team, with technical counsel from the Coordinated CI team, will review allocation requests in years 2, 3, and 4 ($50K per year) for computational requirements that are best suited to the resources or burst access when ARP resources are limited.

0.7.7. Partnerships and Collaborations: DART will utilize a variety of partnerships and collaborative activities to maintain synergy and

relevance in all aspects of the project. Some partners have been identified and we will continue to pursue new partnerships and collaborations during the project.

Letters of Collaboration have been submitted from Arkansas data science companies Stonehenge Technology Labs, Sequoia Southeast, Hytrol Conveyor Company, and Rock Analytics. Stonehenge Labs is a tech company that develops data solutions for logistics and supply chain management sectors. Sequoia Southeast is the asset-based retail arm of Sightline Retail and has logistics and retail sales data at retailers across the US including Walmart, Cost Plus World Market, Costco, Tuesday Morning, Ahold-Delhaize, Amazon, and (soon to be) Safeway/Albertsons. Hytrol Conveyor is a world leader in the design and manufacturing of material handling systems for a variety of industries. Rock Analytics specializes in data visualization services. Each of these has committed to providing additional consultation on industry needs in both the research and education programs. Hytrol, Rock Analytics, and Sequoia Southeast have committed to provide relevant industry problems and data sets for capstone and research projects for DART students, and Stonehenge Labs has committed to hosting two student interns per year. Additional details can be found in their respective letters.

0.7.7.a Research Partnerships and Existing Collaborative Opportunities DART will maintain an ongoing collaboration with the PiLog Group through its academic outreach

organization, the PiLog Academy. PiLog funded two doctoral-level research assistantships (RAs) at UALR last year and is currently funding another RA position. In addition, they are providing the research team with unlimited access to a remote Hadoop Distributed File System (HDFS) Cloudera stack to support research in entity resolution and master data management.


The Sam. M. Walton College of Business (WCOB) at UAF maintains access to Consumer Panel Data from Nielsen at Kilts Center for Marketing and has established relationships allowing access to visits and sales data from Sam’s Club and Dillard’s. WCOB also provides academic access to the Acxiom Infobase Demographical Database including the PersonicX Clusters – Classic, LifeStage Groups, Insurance Groups, Financial Groups, and Digital—as well as population density, ethnic groups and other supporting data used to create the clusters. Researchers also have access to the All-Payer Claims Database (APCD) through the Arkansas Center for Health Improvement and funded by the Arkansas Biosciences Institute.

The J.B. Hunt Innovation Center of Excellence, established in May 2017, is an industry-funded research center at UAF driven through a collaboration between the College of Engineering, WCOB, and J.B. Hunt professionals. The center is funded through a 5-year; $2.75 million grant provided by J.B. Hunt. Research includes working with J.B. Hunt to overcome the challenges associated with developing methodologies that can integrate information of different types and sources to make improved strategic and operational decisions. Successful research projects are measured by their financial impact to J.B. Hunt and the degree in which they disrupt the transportation logistics industry.

0.7.7.b Industry Partnerships and Collaborations: The ACDS is a nonprofit public-private partnership organization that aims to develop, engage, and

retain homegrown top talent in data analytics and computing. The board of ACDS consists of the Blue-Ribbon Committee members, including C-level executives from First Orion, Walmart, EZ Mart, Tyson Foods, Murphy USA, J.B. Hunt, Stephens Inc., Inuvo, AT&T, and others. DART will collaborate closely with ACDS and will receive feedback and advice from the board on numerous project aspects. Additional advisors will be recruited from companies like Acxiom, FIS, Bank OZK, and other large, medium and small businesses in the data sector.

0.8. Overall Risk Management Plan

0.8.1. Challenges and Risks A project of this size and scope faces several challenges and risks including those that were knowable

as the project was developed as well as additional risks and challenges arising from the current global situation. But Covid-19 creates unique and significant challenges to the project. Aside from the added friction of videoconferencing (friction that has lessened to some degree as we’ve learned from our experiences) is the toll taken on faculty and students in time and emotional state. Preparation for and delivery of both online and hybrid classes is far more time consuming and exhausting than all online or all face to face delivery. Mostly importantly, faculty working from home face a complex balance of family and work with children learning from home and spouses working from home all sharing bandwidth and computer resources. Evidence is already mounting that women are more effected by work from home conditions. Schedules and work assignments of all faculty and staff supporting this project have changed since proposal develop – spring, summer, and fall 2020 course delivery moved to remote and hybrid learning, and many project personnel were instructed to work remotely effective March 2020. Out of state travel continues to be suspended from a halt date of March 2020. After an adjustment period, the team effectively shifted to remote working, including successfully executing regular meetings.

Coordinating cyberinfrastructure across seven research campus is primarily a policy and organizational challenge more than a technical challenge. In particular, this project must overcome several organizational issues.

• The lack of a federated identity agreement among participating institutions slows and complicates access to a federated system.


• The need to include campus IT departments who manage the campus enterprise and research networks requires more and larger meetings to agree on approaches that conform to disparate security and enterprise computing policies at each campus.

• Personnel changes at UALR require a shift in visualization strategies to better reflect the current expertise of the team.

• Covid-19 pandemic mitigation measures could challenge our ability to install new equipment at campuses on a timely basis and conduct productive training.

The research goals defined in the project will require the creation of research teams that have not

collaborated before or teams led by excellent, but early career, principal investigators. This poses potential challenges regarding team management and efficient alignment of skills to challenges. While the approaches to development of methodologies dominate a majority of the research in the Learning and Prediction them are understood by participating researchers, data sets to test and validate new methodologies will need to be defined in collaboration with the other themed areas contributed to DART. In addition, this large commitment to research into such a wide variety of complex techniques will require a steady stream of student researchers with quantitative and computational skills. This increased need for graduate students with this expertise may require targeted recruitment over the course of the project.

We see these particular challenges and risk: • Education themes do not have a strong history of close collaboration with the Research

themes. • Data Science capability and infrastructure vary widely within the collaborating institutions. • Enabling technologies are rapidly advancing so participants will need to stay abreast of

current trends. • We are dependent on current industry to provide access to relevant data and other resources

to support the DART program. • Challenges of Academia and Industry working together on real-world data, real-world

problems for class use and training students. • The long timeframe for program approval for public programs at the state level (ADHE) and

the various different agencies that accredit private institutions. • Working with committees at different campuses on course and program approvals. • The lack of human capital to teach courses in data science and computer science at some

collaborating campuses. • The lack of technology infrastructure, both hardware and software, to support the teaching of

data science and computer science courses at some collaborating campuses • Some collaborating institutions have been more involved than others on the ongoing work

that has taken place to build the infrastructure in preparation for this project. • Limited support of international undergraduate students (inability to compensate). • The COVID-19 pandemic limits the opportunity for face-to-face meetings and has led to

travel restrictions. • Support for new and modified programs on collaborating campuses may be adversely

affected by the COVID-19 pandemic. • Participation in workshops and colloquia designed to broaden participation and to develop a

common curricular framework and shared pedagogical strategies may be negatively impacted by the COVID-19 pandemic.

• Recruiting for Educational Activities and Programs may be impacted by the COVID-19 pandemic.

• Attracting Graduate Research Assistants to the program due to the COVID-19 pandemic.


0.8.2. Mitigation Plans: To address these many challenges and risks we propose the following mitigation plans: Campus IT coordination: The first step is to form joint working group that includes together IT

leadership, relevant networking staff, and DevOps staff from UAF and UAMS. Regular monthly meetings, more frequent e-mail exchanges, and shared document repositories will facilitate smooth operation of the core of the CI infrastructure. UALR equivalents will be added to this working group in the second half of year 1 to complete the core team. This core team will develop template CI Plans to share with DART institutions who join the regularly scheduled or special meetings of the working group on an as-needed basis. Getting IT staff involved early in the process and involving as first-class members of the project has proven effective in solving both technical and administrative issue.

Development/integration effort of visualization into thrust areas: Personnel changes at UALR after submission of the EPSCoR DART proposal will require a redefinition of the visualization-support approach in the project. UALR will setup a post-doc position mainly responsible with project-related research and development. This position will be anchored in the EAC and supported/guided by four faculty in the computer-science department with experiences and expertise from a broad spectrum of backgrounds. The faculty together with the post-doc will analyze and discuss current/existing visualization approaches and strategies or the lack thereof with each of the thrust areas as well as with stakeholders involved in the shared testbed development. This will be used to start investigating how existing tools in the thrust areas are already capable of fulfilling needs as well as starting custom developments. Our expectation is that patterns will emerge that should allow to define and create more integrated solutions, which are usable in multiple thrust areas as well as take advantage of the ARCC infrastructure work. The post-doc position will be further augmented by several graduate and undergraduate student positions. UALR is committed in pursuing additional instruments to provide the necessary talent and resources in this area.

Collaboration (esp. in the Covid-19 era): The DART research community needs simultaneous access to a system (Git, Globus, Jupyter notebooks, and others) as well as a shared and familiar video-conferencing platform. Several entities in the project, including AEDC, UAMS and UA, have enterprise licenses to Zoom which are available for DART. Many educational programs have moved on-line or to a hybrid model using Blackboard. ARCC will concentrate initial ARP service roll-out based on highest need across the research community. As noted above, Covid-19 mitigation plans have placed an unusually high load on faculty and, while online meetings and seminars reduce the amount of travel, they can also reduce the effectiveness of the those meetings and limit sidebar conversations that tend to be very productive.

Education and Outreach (esp. in the Covid19-era): In past efforts, the education and workforce components were developed with little or no cooperation between those responsible for the separate components. In this project collaboration and cooperation was built in from the beginning. Eduication CoLeads took part in all the planning sessions, alongside Research CoLead counterparts. The expanding talent pool of highly trained data scientists that are expected to result from the program expansions supported by this project are critical to the ongoing expansion of data science efforts across the state.

The ACDS is also working on workforce and education across the state. The co-leads of the education effort have been working closely with ACDS Director Bill Yoder developing links between industry and education and links between institutions of higher education. Since the Track-1 proposal was submitted, this team has hosted numerous meetings related to the development of a cohesive statewide data science effort. This has given us a greater insight into the challenges we face and gives a better footing on which to build solutions. These efforts give us confidence that we can meet the challenge. As an example, when institutes of higher education (IHEs) talk about shared programs and courses, the participants always bring up obstacles and describe why such collaborations can never happen. As part of our effort to assure


the naysayers, we have already discussed these issues with the ADHE, and we are confident that ADHE will be a contributing partner in our efforts to embed data science into the educational ecosystem of Arkansas. We have similarly engaged industrial partners and we will continue in these efforts. We do not believe that achieving our goals will be easy, but we understand the obstacles and we are committed to overcoming them.

The risks and challenges related to the current global situation were unexpected, and they could make progress more difficult. At this stage there are both known and unknown unknowns. The key factor is clearly the duration of the current global pandemic and whether or not an effective vaccine is developed. Anything that has ongoing negative impacts on IHE enrollments will impact our efforts. However, the emergence of data science provides new opportunities and can provide IHEs expansion possibilities in an area where the global pandemic is not expected to have a negative impact on opportunity. It is clear that an ongoing pandemic will reduce the number of international students.

Communication and Dissemination: Maintaining the campus communication team could be a challenge, it has not been done previously and campuses tend to want to publish research results and new award information as soon as possible. In the past it has been difficult to make sure the grant is cited properly. Our plan to mitigate this is to maintain frequent communication with the comms team and ensure quick turnaround of any needed quotes and citation materials.

The Science Journalism Challenge has never been attempted in Arkansas and is a relatively new concept in general. Possible risks are identifying participants, judges, and ensuring that the winner will be published in a major outlet. Our plan to mitigate this is to establish good relationships with local ad agencies and communications professionals and get their buy-in early on in the SJC development. We will also utilize our existing relationship with Wyoming where a similar effort has been underway for a couple of years, to learn what setbacks they overcame.

1. Coordinated Cyber Infrastructure 1.1. Introduction

The proposed research will be supported by a data science cyberinfrastructure (CI) platform capable of providing secure, distributed, agile, scalable, and on-demand services. We propose to architect and build a private cloud environment, the Arkansas Research Platform (ARP) (Figure 5) and integrate it with existing high-performance computing and petabyte scale storage resources. In combination, these will provide 1) libraries of pre-configured containers designed to support a variety of well-known and novel workflows in machine and statistical learning, graph theory, bioinformatics, and geoinformatics, 2) containers configured for parallel computation and distributed memory on HPC resources for analysis of very large datasets, 3) the ability for researchers to create and share new containers and share, and 4) the ability to stream data to visualization environments both proximate and distant from the computing resources to aid in analysis and meta-analysis of experiments.

1.2. Advancing the State of the Knowledge

A special CI advisory board, chaired by James Deaton, executive director of the Great Plains Network, and composed of the CoPIs of the NSF funded CyberTeam award #1925681, has been formed and will advise in refinement and management of the Arkansas Research Platform described below. This special advisory board will be useful not only in providing external experience in building data science computing platforms, but in coordinating the connection of ARP to the Great Plains Network Research Platform, the Great Plains Augmented Regional Gateway to the Open Science Grid, and on to nationally organized compute and storage resources, complimenting existing connections through XSEDE

A unifying function of the CI is support for the development, optimization and management of analysis pipelines from each of the research themes. Our preliminary experience with this approach has


been quite positive with existing containerized pipelines for image curation, genomics analysis, and machine learning.

1.3. Project Implementation

Goal Goal Name Goal Lead Team Members

CI1 Establish the Arkansas Research Platform as a shared data science resource across the jurisdiction

Cothren, Prior Chaffin, Deaton, Springer, Tarbox

CI2 Visualization for complex data in diverse data-analytics application domains

Springer Conde, Cothren, Huff, Milanova

In May of 2020, UAF and the UAMS entered a partnership to consolidate the management of each

campus’ high-performance computing centers into an entity called the Arkansas Research Computing Collaborative (ARCC). ARCC will manage ARP and all activities related to it. ARCC will be implemented in the first year of the grant and expanded to include resources available at the Emerging Analytics Center at the UALR. These three institutions will act as resource providers and consumers while the other institutions in DART will consume these resources with direct access to big data through Globus, code sharing through GitLab, and computing through interactive and batch sessions on the ARP HPC and private cloud computing resources.

The platform will be configured and managed in accordance with lessons learned from existing similar effort like the Pacific Research Platform and, in particular, the Great Plains Research Platform; and with input from an advisory board chaired by the executive director of the Great Plains Network, James Deaton. We will use best practices learned from these projects, with the goal to enable sharing resources among topic-integrated, but geographically diverse, research teams. Remote storage and processing of larger compute jobs will be hosted at XSEDE national computing sites and commercial providers (e.g., Google, Azure). Connectivity across the state will be provided by ARE-ON and serve the seven research campuses and two education-focused HBCU campuses, Philander Smith College and Shorter College.

ARP will use Globus Research Data Management services for big data sharing and management across the seven research campuses. Endpoints will be set up at UAF, UALR, UAMS, ASU, UCA, SAU, and UAPB. Data will be backed up to the NSF-funded regional tape resource OURRstore at Oklahoma University. Code archiving and sharing will use the existing GitLab for Enterprise installation at UA. ARP will support a variety of research computing platforms: traditional bare-metal HPC jobs, Singularity containers, and kernel virtual machines (KVMs) on the existing Pinnacle and Grace clusters as well as the new data science cluster funded as part of this proposal. The Sample Linux Utility for Resource Management (SLURM) scheduler will be used to provision all three types of jobs. Singularity containers are a widely accepted, secure standard in a multi-user HPC environment where access to the data of other users must be restricted. The container jobs will require a user to either download a container from external repositories like NVidia NGC (in native Singularity format) or Docker Hub (easily convertible to Singularity format) or use containers stored on shared local storage on Pinnacle or Grace. Once the container is in place, a single-line command in the SLURM job script will bind the user's input data directory to the container and run the executable inside the container on the input data. The job terminates when either the executable in the container finishes processing the input, or the job exceeds the requested wall time. A variety of big data management resources such as HDFS and Apache Spark will be enabled using the MapR Sandbox and internet-facing graphical interfaces will be provided to run services such as Jupyter Notebook, RStudio Server, and the HPC scheduler, Open OnDemand, at both UAF and UAMS.


Figure 5: The computational backbone of ARP consisting of existing and new, grant-supported equipment.

To support the proposed research, a dedicated data science-oriented cluster will be purchased and physically joined to Pinnacle at UAF. It will consist of approximately 46 nodes each with dual-Xeon Cascade Lade 20-core processors, 768 GB of memory, 480 GB local solid-state storage, one Nvidia Tesla V100 GPU. Nodes will be connected via EDR or HDR InfiniBand. Additional storage (500TB) will be added at UAMS and 48 nodes from the existing Grace cluster re-tasked to contribute to ARP. To be investigated is the dynamic allocation of resources between ARP and traditional local HPC based on demand. Currently, only UAF has 100Gb/s capability to connect to the ARE-ON backbone. Proposed funds will add 100Gb/s capability to the UAMS computing center in year 2, finally realizing a high-bandwidth connection between the two major computing clusters in the state via links provided by the statewide ARE-ON backbone.

ARP will be connected with the Great Plains Network Research Platform, patterned after the Pacific Research Platform, leveraging regional resources to potentially increase the compute infrastructure available to DART and other researchers in Arkansas. ARP plans to connect Arkansas-based resources to the Open Science Grid (OSG) using gateway nodes, leveraging work being done for the CC* Compute award #2018766: GP-ARGO: The Great Plains Augmented Regional Gateway to the Open Science Grid. The connection will give DART and other researchers in Arkansas access to the vast resources for High Throughput Computing available on the Open Science Grid, while providing resources to the Open Science Grid that are not being used for local projects. ARP will partner with The Carpentries to deliver high quality data science-oriented training in scientific software development, data management, and code management. UA is currently a Silver Member and will seek to offer at least 5 online training sessions per year in various locations.


Goal 1.1 (CI1) Establish the Arkansas Research Platform as a shared data science resource across the jurisdiction Objective 1.1.a: Establish the Arkansas Research Computing Collaborative (ARCC) Objective 1.1.b: Upgrade cluster for data science research activity and integrate with existing resources Objective 1.1.c: Establish a science DMZ in Little Rock (UAMS, UALR) and high-speed connection with UAMS Objective 1.1.d: Establish a data and code sharing environment (GitLab and Globus) Objective 1.1.e: Establish necessary controls to store and manage controlled unclassified, HIPAA-related, and proprietary information at UA and UAMS (other institutions if possible)

Goal 1.1 Output Metrics Hardware and software infrastructure improvements (5): -- Install, configure, and make available data science nodes on Pinnacle Portal -- ScienceDMZ at UAF and UAMS/UALR -- 100Gb connection between ScienceDMZ -- Establish a dedicated DART GitLab repository -- Setup Globus data management services to point at DART storage arrays

Documentation and user guides (4): -- Create one (1) ARCC technical management document -- Amend existing MOU for ARCC expansion -- Create two (2) CI Plans (1 x UAF, 1 x UAMS) -- Create one (1) GitLab user guidelines reference document

Workshops, demonstrations, and trainings (38): -- Host two (2) online workshops per year for onboarding to ARP resources in YR2-5 (8 total) -- Host five (5) online software carpentry workshops per year in YR2-5 (20 total) -- Train and certify two (2) new software carpentry instructors per year in YR2-5 (10 total)

Applications and platforms (5): -- Create one (1) distributed computing testbed for HDFS, Apache Spark, others (DC) --Create four (4) spatiotemporal testbeds for (CI/DC/SM/LP)


Objective 1.1.a Specific Milestones

Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Create ARCC advisory board with regional partners (GPN)

Establish CI advisory board

Activity 2: Establish ARCC governance, operations, and staff between UA and UAMS

Create a document defining organizational structure, roles, and responsibilities of ARCC

Activity 3: Expand ARCC to include UALR as a provider and other DART participants as consumers

Amend existing MOU for ARCC expansion

Activity 4: Create UAF CI Plan to support DART (prior to 1.1.b and 1.1.c)

Create UAF CI Plan

Objective 1.1.b Specific Milestones

Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Specify and purchase data science cluster based on document from 1.1.a

Issue UA purchase order for additional equipment

Receive data science nodes for Pinnacle (anticipated)

Activity 2: Test and deploy hardware elements for Pinnacle expansion for DART

Install, configure, and make available data science nodes on Pinnacle

Activity 3: Install and configure data science cluster to work with existing resources at UA, UAMS, UALR resources

Collect testbed specifications and software/platform needs

Create containerized Hadoop-based testbed for DC

Create containerized testbeds for SA and SM

Create additional containerized testbeds for SA and SM


Objective 1.1.c Specific Milestones

Year 1 Year 2 Year 3 Year 4 Year 5

Activity 1: Specify and purchase 100Gb switch

Issue UAMS purchase order for 100 Gb switch

Receive 100 Gb switch (anticipated)

Activity 2: Install 100Gb switch

Install and configure new 100 Gb switch

Activity 3: Establish ScienceDMZ at UAMS

Create UAMS CI Plan Specify and acquire additional DMZ components

Establish and validate 100Gb link to UAF and integrated DMZ

Objective 1.1.d Specific Milestones

Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Create/identify federated identify or other authentication mechanism for all sites that provides access to core ARP resources

Establish federated ID for all project participants

Activity 2: Setup dedicated GitLab repository

Create and publish document outlining GitLab user guidelines, minimum standards for code repositories, and best practices.

Activity 3: Setup Globus Data Management Services

Globus Data Management contract executed

Establish Globus endpoints to DART storage arrays at key sites

Activity 4: Engage other research themes to develop research-specific training modules in e.g. Python, R, Git, HPC, Singularity

-- Host 5 online software carpentry workshops -- Train 2 software carpentry instructors





Activity 5: Develop and deploy training materials for code sharing, large data transfer protocols

Host 2 online ARP-specific training sessions




Objective 1.1.e Specific Milestones

Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Identify the number and type (HIPAA, proprietary economic, CUI, etc.) of private and secure data sources that will be need to be accessed by DART researchers.

Collect research theme needs

Activity 2: Setup capacity for storing and managing CUI and HIPAA data at UAF

Deploy restricted access storage

Draft user guidelines and policy


Goal 1.2 (CI2) Visualization for complex data in diverse data-analytics application domains Objective 1.2.a: Investigate state-of-the-art visualization solutions Objective 1.2.b: Define domain-specific integration of visualization solutions Objective 1.2.c: Introduce/integrate visualization for shared test beds

Goal 1.2 Output Metrics Publications, presentations, and reports (3): -- Three (3) presentations, reports, or other publications: 1 in YR1 and 2 in YR2

Workshops, demonstrations, and trainings (4): -- One (1) online workshops per year for advanced visualization in YR2-5 (4 total)

Applications and platforms (9): -- Develop one (1) visualization solution for each research theme, including CI (5 total) -- Integrate one (1) visualization into existing testbed for each research theme (4 total)


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Investigate/define state-of-the-art visualization

1 presentation/report

Activity 2: Investigate standard tools for data science visualization


Activity 3: Investigate/define data exchange strategies and their relationship to other research themes



Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Develop and deploy visualization infrastructure software

Collect research theme needs

1 software prototype (alpha)

1 software prototype (beta)

Activity 2: Develop domain-specific visualization solution for DC

1 software prototype

(alpha) 1 software prototype

(beta)


Activity 3: Develop domain-specific visualization solution for SA



(beta)

Activity 4: Develop domain-specific visualization solution for SM



(beta)

Activity 5: Develop domain-specific visualization solution for LP



(beta)


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Integrate visualization into existing testbeds for automated data curation environment DC/SM



(beta)

Activity 2: Integrate visualization into existing testbeds for social media-linked, GIS platform CI/DC/SM/LP



(beta)

Activity 3: Integrate visualization into existing testbeds for bioinformatics workflows DC/SM



(beta)

Activity 4: Integrate visualization into existing transaction-based testbed LP/DC



(beta)

Activity 5: Engage other research themes to develop research-specific advanced visualization training

Host 1 online advanced

visualization workshops








2. Data Life Cycle and Curation

2.1. Introduction

The overall goal of the Data Life Cycle and Curation Theme is to create unsupervised and scalable methods that significantly increase the level of automation in the data curation process from acquisition to disposal. The primary curation processes targeted for automation in DART research are data quality management, data integration, and data governance. While there are many tools already available for these processes, they are all dependent upon human supervision to work effectively. One of the most common complaints of data scientists is they only spend 20% of their time on modeling and problem solving and the other 80% on data preparation. The same is true in industry where most of the effort in data operations is consumed in cleansing, standardizing, and integrating data prior to its actual application in information products. As data volume continues to grow at a rapid rate, the curation process has become a significant bottleneck for data operations resulting in long delays before data are available for analytics and other data-driven operations.

The Arkansas Science and Technology Plan which has three primary objectives. 1. To identify opportunities for academic and industrial collaboration 2. To align future investments in university research competencies with industry areas of

technology focus 3. To stimulate improvement in technology skills and talent development. Because the lack of automation in data curation is a problem for both industry and academic

research, the theme of automating data curation fits well with the first and second objectives. Collaborating with industry through testing real-world datasets will be an essential component of the research. As the first university research to focus on this research theme, the researchers, including student research assistants at participating schools, will develop high-level skills in data analytics, data governance, and machine learning.

Within Arkansas, the UAMS, UAF, and UALR campuses will be the primary drivers for the research. UAMS will bring into the research the special needs of data curation for biomedical informatics, UAF expertise in AI and machine learning, and UALR the industry perspective.

In addition to the internal partners, the preliminary research has already attracted interest from a number of external academic and industry collaborators including the MIT Chief Data Officer and Information Quality program, PiLog Group, and Noetic Partners. As the research matures it is likely to attract more collaborators and potentially result in the development new open source and commercial products and generate new business opportunities.

2.2. Advancing the State of the Knowledge:

The three most time-consuming data preparation processes are data cleaning, data integration, and data tracking (data governance). The vision for the research is a “data washing machine.” People are accustomed to throwing their dirty laundry into the washer along with some soap, setting the dials for the type of clothes, and letting the washer operate automatically. A data washing machine would work in a similar manner on dirty data - simply ‘throw in dirty data’, push a button, and out comes ‘clean’ or curated data.

If such machines can be built, the benefits are enormous. They will revolutionize data operations in research, industry, and government. When data cleaning, data integration, and data governance become unsupervised, automated processes, then much more data can be ingested and analyzed, and greater advances in data analytics can be made is less time. At the same time, the improvements in data governance will make enterprise data assets more secure while making them more available and discoverable for authorized users.




DC1 Automate heterogeneous data curation Talburt

Talburt, Cothren, Liao, Liu, Rainwater, Tudoreanu, Ussery, Wang, Xu, Yang

DC2 Explore secure and private distributed data management

Talburt Talburt, Wang, Tudoreanu, Pierce, Liu, Rainwater

DC3 Harmonize multi-organizational and siloed data

Ussery Ussery, Byrum, Jun, Yang, Liu, Rainwater

The initial research will focus on three important use cases from industry and academia. The first use

case is “multiple sources of the same information.” Solving this use case is Objective 2.1 of the Activity Matrix. In Journey to Data Quality (Lee, et al., MIT Press, 2006), having multiple sources is at the top of the list of the 10 most common root causes of data quality problems. It is a process repeated over and over every day in organizations across the globe. They receive large files of information about customers, patients, products, events, and other entities from multiple sources in different formats. The current approach is for a data analyst to profile each source and design and ETL process to transform it into a standard layout. After each source is prepared the records then go through a linking process designed by a data integration analyst so that records describing the same entity can be identified, linked together, and eventually integrated into an information product. The goal of this use case is to create an unsupervised process to produce equivalent results. There are four concurrent streams of work within this first use case (DC Goal 1). These are:

• Automate Data Quality Assessment, Years 1 - 5 • Automate Data Cleansing, Years 1 - 5 • Automate Data Integration, Years 1 - 5 • Implement Data Cleansing and Data Integration Models in HDFS, Years 2 - 5 The second use case is around the design of an enterprise positive data control (PDC) system. As

more and more organizations recognize data and information as a key asset, they are adopting the discipline of data governance. One of the key objectives of data governance (DG) is to always know exactly what data is present in the system, exactly where it is located, and what it represents. Just as with data curation, DG requires a great deal effort by a large number of persons in the role of “data stewards” making it difficult to keep up with the growing data volume and variety. At the same time, an organization must comply with a growing number of new data regulations such at the General Data Protection Regulation (GDPR) from the European Union, and the California Consumer Protection Act (CCPA). Both regulations allow a consumer the “right to be forgotten,” i.e. having their personal identifying information (PII) deleted from all systems. Without effective DG, organizations are at risk of non-compliance carrying substantial fines and penalties. The goal of this use case is to create a controlled data operations environment, in which all data access is controlled and every data action is automatically tracked and recorded. There are three concurrent streams of work within this second use case (DC Goal 2). These are:

1. Build a POC for Positive Data Control (PDC), Years 1 - 3 2. Extend the Functionality of the POC, Years 3 - 4 3. Implement Data Portability and Exchange Functionality


Figure 6: data washing machine

The third use case focuses on the curation of genomic and proteomic data, which is now growing at the incredible rate of petabytes of new data per week. The goal of this use case is to design scaffolding systems that capture and disseminate both data and processing steps, and that can function as a democratizing tool for data in a similar manner as blockchain functions for currency. This effort intends to allow sharing of data processing steps that can be beneficial to (a) valid integration of data from multiple silos, (b) multi-enterprise, distributed data governance and curation efforts, (c) supporting an organization that has multiple or lax data governance processes, and (d) upfront automated data curation by machine learning algorithms. There are three concurrent streams of work within this second use case(DC Goal 3). These are:

• Standardize pipelines for genome and proteome storage, retrieval, and visualization, Years 1 - 2 • Automate quality scores for biological sequence data, Years 2 - 3 • Apply machine learning methods to systems biology, Years 3 – 5


Goal 2.1 (DC1) Automate heterogeneous data curation Objective 2.1.a: Automate Reference Clustering / Automate Data Quality Assessment Objective 2.1.b: Automate Data Cleansing Objective 2.1.c: Automate Data Integration

Goal 2.1 Output Metrics Publications, presentations, and reports (46): -- 14 research publications describing new methods and processes -- 11 journal and conference publications -- 16 presentation -- Five (5) PhD dissertations

Patents and start-ups (2): -- Two (2) potential patents and/or business incubation

Workshops, demonstrations, and trainings (5): -- Five (5) conference workshops

Datasets and algorithms (18): -- Nine (9) Mathematical formulations and/or algorithms for optimal collaborative data collection and cleansing. -- Seven (7) Mathematical formulations, algorithms, and/or statistical methods for need- and prediction-based data collection, and scalable decision making



Activity 1: Define metrics for data quality to measure impact of unsupervised data cleansing on data standardization and reference clustering.

-- Define at least one metric for completeness, standardization, and clustering quality of unstandardized reference data -- Design and implement an unsupervised algorithm for each metric

Design and implement an algorithm using ML or Graph techniques for one metric

Design and implement a scalable algorithm in HDFS for one metric

Design and implement a scalable algorithm in HDFS for all metrics


Activity 2: Set baseline data quality for initial test datasets used in prior research and acquire additional test datasets

-- Establish baseline quality using supervised methods for existing datasets -- Compare results of unsupervised quality metrics developed in Activity 1 to supervised results

-- Add 5 new person and 5 new business reference datasets for testing, at least 2 real-world -- Add 5 new product reference datasets

Add 3 new person and 2 new business reference datasets with more than 1 million records for testing HDFS code

Add 5 new product reference datasets with more than 1 million records for testing HDFS code

Activity 3: Curate test datasets and make available to other researchers

Establish a repository for the reference datasets and make available to other researchers

Add new reference datasets to the repository, as needed

Activity 4: Develop a framework for collaborative data collection and cleansing for knowledge discovery

Formulate a hierarchical and as-needed data collection and cleansing strategy

-- Refine the formulation by including various practical constraints and test on small-scale problems -- Formulate a collaborative data collection strategy involving multiple teams

Solve large-scale problems by considering a tree-based tool for data clustering and cleansing

Solve large-scale problems considering hierarchical and unsynchronized data collection, and test on inland waterway datasets

Activity 5: Develop a need- and prediction-based feedback mechanism for future data collection and making scalable decisions

Formulate a framework for sequential data collection on an as-needed basis Refine the formulation by including various practical constraints and test on small-scale problems

-- Formulate a Bayesian framework for sequential data collection based on predictive models -- Investigate analytical approaches for using large datasets for different levels of decision making

Refine the Bayesian framework for sequential data collection based on predictive models and medium-sized data

Study an optimum data guided approach for sequential learning and dynamic data collection

Solve large-scale sequential data analysis and multi-level decision making problems and test on inland waterway datasets



Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Improve the unsupervised frequency-based data cleansing method used in prior POC; Explore and test alternative methods and models for unsupervised data cleansing including ML, AI, and graph approaches

-- Document and train team on data cleansing methods developed in prior research -- Design and implement in Python or Java improvements to the prior frequency-based approach

-- Design and test an ML or Graph implementation to the prior frequency-based approach -- Design and test new techniques for unsupervised data cleansing in Python, Java, ML, or Graph

Continue to design and test new techniques for unsupervised data cleansing in Python, Java, ML, or Graph

Activity 2: Migrate successful data cleansing models into a scalable HDFS processes

-- Refactor and migrate prior frequency-based approach into a scalable HDFS process

Refactor and migrate most successful of new data cleansing techniques into a scalable HDFS processes


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Improve the unsupervised frequency-based data integration method used in prior POC and explore and test alternative methods and models for unsupervised data integration including ML, AI, and graph approaches

-- Document and train team on reference clustering method developed in prior research -- Design and implement in Python or Java improvements to the prior frequency-based approach

-- Design and test an ML or Graph implementation to the prior frequency-based approach -- Design and test new techniques for unsupervised data cleansing in Python, Java, ML, or Graph




Activity 2: Migrate successful reference clustering models into a scalable HDFS processes

Refactor and migrate prior frequency-based approach into a scalable HDFS process

Refactor and migrate most successful of new reference clustering techniques into a scalable HDFS processes



Goal 2.2 (DC2) Explore secure and private distributed data management Objective 2.2: Build a POC and demo for Positive Data Control (PDC)

Goal 2.2 Output Metrics Publications, presentations, and reports (6): -- Three (3) conference presentations -- Three (3) research papers

Workshops, demonstrations, and trainings (3): -- Three (3) demonstrations of data exchange between PDC Systems

Applications and platforms (2): -- One (1) working prototype of a positive data control system -- One (1) novel design of a system to control data access and movement in synchronization with a data catalog/governance standards



Activity 1: Build a POC and demonstration code for a Positive Data Control system layer forcing all the tools read/write operations to synchronize with the platforms metadata tool

-- Setup a test HDFS platform with at least one processing function (e.g Hive), metadata function (e.g Atlas), and security function (e.g Ranger); -- Build POC with a simple PDC layer where Hive user is forced to go through PDC layer for all read/write operations

Modify POC to synchronize Hive operations with metadata layer (Atlas) and security permissions (Ranger)

Add APIs to PDC layer to control an additional data transform and one data movement tools

Create and give demo of typical use case where PDC controls all user actions


Activity 2: Data exchange between PDC systems

Clone PDC system to create second PDC system. Add code to PDC layer to import/export datasets conforming to ISO 8000-110

Demo exchange of data between the two PDC systems with automatic update of metadata in receiving system

Goal 2.3 (D3) Harmonize multi-organizational and siloed data Objective 2.3.a: Standardize pipelines for genome and proteome storage, retrieval, and visualization Objective 2.3.b: Automate quality scores for biological sequence data Objective 2.3.c: Apply machine learning methods to systems biology

Goal 2.3 Output Metrics Publications, presentations, and reports (17): -- Five (5) papers published on standardized pipelines for genomics and proteomics. -- Five (5) conference presentations -- Five (5) Journal publications -- Two (2) PhD dissertations

Workshops, demonstrations, and trainings (1): -- Host one (1) workshop on how to use these standardized pipelines

Datasets and algorithms (2): -- At least one (1) standardized database with genomics and proteomics quality scores shared with DART researchers -- One (1) algorithm (code with associated training and testing data)



Activity 1: Define and download datasets to be curated

Build genomics database, including quality scores, and gene/ protein annotation

Extend to proteomics database - all for fast characterization of proteins (links to SwissProt)

Update databases (every 6 months)

Activity 2: Optimize data storage and retrieval

Use Elastic Cloud Storage for fast retrieval

Develop integrated database for proteomics & genomics, including annotations

Update databases (every 6 months)


Activity 3: Develop visualization methods

Prototype of R-BioTools for visualizing genomes

Publish one (1) R-BioTools paper for visualizing genomes

Develop visualization methods for very large trees

Visualization of phylogeny for Ark. Pathogens



Activity 1: Develop pan-genome and Pan-proteome databases

Develop architecture / structure for rapid storage/retrieval of taxa-specific pan- and core-genomes

Develop rapid storage/retrieval for core- and pan-proteomes

Activity 2: Develop taxonomy links to downloaded genomes/proteomes

Compare duplicate, known type strain genomes using ANI, Mash, 16S rRNA

Use Mash and other methods to assign nearest neighbors in phylogenetic space.

Update taxonomy (every 3 to 6 months)

Activity 3: develop a genomic database for Arkansas genomic pathogen surveillance of antimicrobial resistance

Build Arkansas pathogen database, with links to known pathogens; develop GPU-based methods for fast calculation of genomic distances

Finish development of GPU-based methods for fast calculation of genomic distances

Publish one (1) taxonomy paper


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Define training sets to be used for ML

Identify key datasets and problems for ML

Develop ML models for known toxins

Develop ML models for antibiotic resistance

Apply ML models to Ark. Pathogens

Activity 2: Integrate multi-omic models for ML

Integrate genomic / microbiome / taxonomy datasets (petabytes)

Integrate genomic, transcriptomic, proteomic, and metabolomic datasets (petabytes)

Integrate model organisms (mouse, rat, human, yeast, etc.) as well as microbial

Publish one (1) Integration paper

Activity 3: Benchmark ML results

Develop Benchmarking Standards for ML of pathogens

Publish one (1) ML Benchmarking papers


3. Social Awareness

3.1. Introduction (or Preface)

Social awareness is pivotal for those who work with data analytics and is a key factor that affects the uses, benefits, and risks of big data. It is a common practice for both government agencies and private entities to collect and integrate large amounts of many different kinds of data, process it in real time, and deliver the product or service to consumers. There are increasing worries that both the acquisition and subsequent application of big data analytics could cause various privacy breaches, render security concerns, enable discrimination, and negatively affect diversity in our society. All these concerns affect public trust regarding big data analytics and the ability of institutions to safeguard against such negative social outcomes. As such, social awareness should be an integral part of research and training in the area of data analytics. In this theme, we are focused on developing cutting-edge socially aware data analytics to address social concerns and meet laws and regulations in national-priority applications, thus better enabling big data analytics to promote social good and prevent social harm.

Our major research goals are to develop novel techniques to provide privacy preservation, fairness, safety, and robustness to a variety of data analytics and learning algorithms including automated data curation, social media and network analysis, and deep learning, and ensure the adoption of the developed techniques meet regulations, laws and user expectations.


Our developed technology can achieve meaningful and rigorous privacy protection when mining private data or collecting sensitive data from individuals; ensure non-discrimination, due process, and understandability in decision-making; achieve safe adoption, and robustness of machine learning and big data analytics techniques, especially in adversarial settings; and help incorporate social awareness in domain- or application-specific projects. Our research projects in this theme will advance the state of the knowledge in the following perspectives.

SA1: Privacy-preserving and attack resilient deep learning • Vulnerabilities of deep learning algorithms under existing and new attacking models • A universal threat- and privacy-aware deep learning framework to achieve meaningful and

rigorous differential privacy protection and resilience against various attacks • Better understanding of the applicability of threat- and privacy-aware deep learning models in

real world applications SA2: Socially aware crowdsourcing • Improve crowdsourcing data quality with considerations of uncertainty • Enhance available inference and learning models with novel algorithms for improved

effectiveness and efficiency • Verify and validate the robustness and trustworthiness of information from crowdsourcing data SA3: User-centric data sharing in cyberspaces • Understand personal identifying information and their privacy issues • New appropriate multimodal deep learning techniques to identify discriminative and

stigmatizing information • A user-centric privacy monitoring and protection framework SA4: Deep learning for preventing cross-media discrimination • Deep learning-based techniques to detect cross-media discrimination • New generative adversarial models to remove cross-media discrimination • A joint multi-modal deep learning framework to detect and prevent cross-media discrimination.

Test and evaluate the proposed techniques and models with large-scale social media data SA5: Marketing strategy design with fairness


• Text mining and sentiment analysis of user-generated data from social media and consumer shopping records to extract customer-desired product features

• Network-based modeling of customer preference incorporating marketing parameters • Design of marketing strategies with fairness consideration and validate the approach SA6: Privacy-preserving analytics in health and genomics • Machine learning algorithms and software, and advanced security and privacy technologies, for

privacy-preserving data analytics • Innovative frameworks and practical privacy-preserving techniques for controlled genomics and

health data • New algorithms and technologies to work with high-dimensional heterogeneous data sources

and bioinformatics workflows into the private cloud environment, the ARP SA7: Cryptography-assisted secure and privacy-preserving learning • Privacy-preserving federated learning methods through combining cryptography techniques and

privacy models • New algorithms to protect the privacy of classification input data from the server hosting

machine learning models • New techniques for assessing and protecting the trustworthiness of training data and machine

learning models



SA1 Privacy-Preserving and Attack Resilient Deep Learning X. Wu Q. Li, Zajiciek

SA2 Socially Aware Crowdsourcing Hu N. Wu, X. Wu

SA3 User-centric Data Sharing in Cyberspaces

N. Wu Q. Li, Hu

SA4 Deep Learning for Preventing Cross-Media Discrimination

L. Zhang X. Wu, Sha, Zajicek

SA5 Marketing Strategy Design with Fairness Sha L. Zhang, X. Wu

SA6 Privacy-Preserving Analytics in Health and Genomics Huang

Q. Li, M. Yang, Pierce, Ussery, CI team

SA7 Cryptography-Assisted Secure and Privacy-Preserving Learning Q. Li Huang, N. Wu

The research goals of the Social Awareness research theme are to develop novel techniques to

provide privacy preservation, fairness, safety, and robustness to a variety of data analytics and learning algorithms including automated data curation, social media and network analysis, and deep learning, and ensure the adoption of the developed techniques meet regulations, laws and user expectations.

The theme team will achieve the above goals through working on seven research projects under this theme and collaborating with other theme teams. The Social Awareness Activity Matrix presents details of objectives, activities and planned milestones over the five-year period for each of the seven projects.


Goal 3.1 (SA1) Privacy-Preserving and Attack Resilient Deep Learning Objective 3.1.a: Identify potential vulnerabilities of deep learning algorithms Objective 3.1.b: Develop a universal threat- and privacy-aware deep learning framework Objective 3.1.c: Conduct comprehensive evaluations of the proposed framework and models

Goal 3.1 Output Metrics Publications, presentations, and reports (5): -- Two (2) conference papers -- One (1) journal paper -- One (1) thesis -- One (1) proposal

Workshops, demonstrations, and trainings (1): -- One (1) tutorial given at major AI conference


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Research existing attacks including model inversion attacks and data poisoning attacks and capture mechanisms behind the threat models

Document literature research of attack models and mechanisms behind attacks

Activity 2: Study the potential risks due to correlations among input data features, parameters, output, target victims, and latent feature space in deep learning algorithms

Initiate theoretical investigation on the risks of deep learning algorithms

Disseminate the findings of both theoretical and empirical studies on risks of deep learning algorithms


Activity 3: Study the sensitivity and impact of input data features, parameters, and the objective functions on the model output and identify appropriate differential privacy preserving mechanisms for different computational components in a variety of deep learning models

Initiate theoretical investigation of privacy preserving mechanisms.

Disseminate the findings of both theoretical and empirical studies on privacy preserving mechanisms used for deep learning algorithms


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Investigate the tradeoff of achieving privacy, resilience to adversarial attacks, and utility

Research the tradeoff of privacy, resilience, and utility

Complete both theoretical and empirical studies on privacy, resilience, utility tradeoff in the deep learning setting

Activity 2: Study the mechanisms of redistributing injected noise across input data features, model parameters, and coefficients of objective functions based on their vulnerability and impact on the model output

Examine the noise redistribution mechanism

Complete both theoretical and empirical studies on the noise redistribution mechanism in the deep learning setting

Activity 3: Develop and implement threat- and privacy-aware deep learning models

Design algorithms of threat- and privacy-aware deep learning models

Complete initial implementation

Complete the model improvement and extension




Activity 1: Evaluate the developed framework and models against baselines using benchmark datasets

Complete the data collection of benchmark datasets and implementation of baselines

Complete empirical evaluation of developed models against baselines on benchmark datasets

Activity 2: Evaluation and validation with participating companies

Complete the data collection (including preprocessing) of real-world datasets

Complete empirical evaluation of developed models against baselines on collected real world datasets

Goal 3.2 (SA2) Socially Aware Crowdsourcing Objective 3.2.a: Improve crowdsourcing data quality with considerations of uncertainty Objective 3.2.b: Enhance available inference and learning models with novel algorithms for improved effectiveness and efficiency Objective 3.2.c: Verify and validate the robustness and trustworthiness of information from crowdsourcing data



Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Allow uncertain labels in crowdsourcing data collection

Selected the approaches through literature review

Implemented and tested

Activity 2: Aggregate raw labels after label collection

Computational schemes are identified

Implemented and tested

Activity 3: Filter out possible noises to further improve data quality

Identified possible sources of noises

Filtering algorithms designed

Algorithms implemented and tested




Activity 1: Build theoretic foundations

Specified mathematical requirements

Activity 2: Develop learning models and inference algorithms

Algorithms designed to meet specification


Activity 3: Test and apply these learning models and algorithms

Testing dataset selected

Initial tests completed Refined learning models and algorithms


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Establish additional evaluation metrics

Quality metrics established

Activity 2: Develop algorithms to calculate the metrics

Quality metrics are quantified


Disseminate the findings of the algorithms and refinements

Activity 3: Verify and validate computational results

Completed system integration and testing

Performance evaluated and compared

Goal 3.3 (SA3) User-centric Data Sharing in Cyberspaces Objective 3.3.a: Investigate on personal identifying information and their privacy issues Objective 3.3.b: Investigate appropriate multimodal deep learning techniques to identify discriminative and stigmatizing information Objective 3.3.c: Develop a user-centric privacy monitoring and protection framework





Activity 1: Research state-of-art entity identification techniques for non-structure data

Document and disseminate the findings on personal identifying information and their privacy issues

Activity 2: Investigate appropriate techniques for identifying context-aware sensitive information

Identify and disseminate the findings on the sensitivity of information in different context

Identify appropriate techniques for identifying context aware sensitive information.

Activity 3: Develop appropriate text analysis techniques to identify sensitive information from unstructured data

research appropriate text analysis techniques to identify sensitive information from unstructured data

Develop appropriate techniques for identifying sensitive information from unstructured data



Activity 1: Research state-of-art multimodal deep learning techniques for identifying private sensitive information

Study and document state of art multimodal deep learning techniques

Document and disseminate the findings on the determination of appropriate multimodal techniques for detecting sensitive information


Activity 2: Investigate appropriate techniques for identifying discriminating and stigmatizing information

Document and disseminate the findings of state-of-art techniques for identifying discrimination information

Document and disseminate the findings on the determination and development of appropriate techniques for identifying stigmatizing information

Activity 3: Develop appropriate deep learning text analysis techniques to accurately remove discriminating and stigmatizing information

Design deep learning techniques for removing discriminating and stigmatizing information Implement deep learning techniques for removing discriminating and stigmatizing information

Test and document the efficiency of the techniques and improve them when necessary


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Develop appropriate techniques for monitoring personal information disclosure on the Internet such as those from government records, news reports, and online documents

Research the key ventures for sharing/publishing/releasing PII

Define and document appropriate techniques for monitoring PII disclosure on the Internet


Activity 2: Develop a risk assessment method for possible privacy breach given the amount of personal identifying information disclosed/published

Research the relationship among PII attributes and how the release of one attribute would affect the overall privacy

Develop a risk assessment framework for assessing the possible risk of privacy breach given the amount of PII has been released

Document the findings from the testing and refining of the framework

Activity 3: Develop appropriate techniques for safeguarding sensitive information by helping end users monitor and proactively control the release of their personal information.

Develop appropriate techniques for safeguarding sensitive information based on the research from activity 1 and 2

Document the findings from the resting and refining of the framework

Goal 3.4 (SA4) Deep Learning for Preventing Cross-Media Discrimination Objective 3.4.a: Explore deep learning-based techniques to detect cross-media discrimination. Objective 3.4.b: Design generative adversarial models to remove cross-media discrimination. Objective 3.4.c: Develop a joint multi-modal deep learning framework to detect and prevent cross-media discrimination. Test and evaluate the proposed techniques and models with large-scale social media data.


Workshops, demonstrations, and trainings (1): -- One (1) tutorial given at major AI conference

Datasets and algorithms (1): -- One (1) algorithm (code with associated training and testing data)



Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Use deep convolutional neural networks (CNN) to recognize discrimination-sensitive objects from images.

Initiate theoretical investigation on using CNN to recognize discriminatory objects

Complete design and implementation of the CNN-based model

Activity 2: Adopt long short-term memory (LSTM) network to model the text.

Initiate theoretical investigation on using LSTM to model discriminatory text

Complete design and implementation of the LSTM-based model

Activity 3: Utilize bilinear model to capture the implicit relationship between the detected discrimination-related objects and the text.

Initiate the theoretical investigation on the implicit relationship between the detected discrimination-related objects and the text

Complete design and implementation of the bilinear model


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Adopt mixture Generative Adversarial Nets framework for generating perceptually similar and discrimination-free image patches.

Initiate the theoretical investigation on using GAN to generate discrimination-free image patches

Complete design and implementation of the GAN model

Activity 2: Applies the encoder-decoder mechanism to automatically generate the discrimination-free text based on the original text.

Complete preliminary theoretical investigation on using encoder-decoder to generate discrimination-free text

Complete design and implementation of the encoder-decoder model


Objective 3.4.c Specific Milestones Year 1 Year 2 Year 3 Year 4 Year 5

Activity 1: Develop a comprehensive multi-modal deep learning framework for jointly learning from both images and text to detect and prevent cross-media discrimination.

Initiate the theoretical investigation on the comprehensive multi-modal deep learning framework

Complete model design and implementation

Activity 2: Test and evaluate the proposed techniques and models from available data sources in social networks like Facebook, Instagram, and Foursquare.

Complete social media data collection. Compete evaluation of CNN and LSTM models

Complete evaluation of the bilinear model

Complete evaluation of GAN and encoder-decoder models

Complete evaluation of the comprehensive multi-modal deep learning framework

Goal 3.5 (SA5) Marketing Strategy Design with Fairness Objective 3.5.a: Text mining and sentiment analysis of user-generated data from social media and consumer shopping records to extract customer-desired product features Objective 3.5.b: Network-based modeling of customer preference incorporating marketing parameters Objective 3.5.c: Design of marketing strategies with fairness consideration and validate the approach






Activity 1: Data collection from social media data, e.g., Amazon and Facebook, using vacuum product as the application context

Document and disseminate the findings of literature research and evaluation of the target products for case study

Document and disseminate the findings of data collection from social media including both review data and production information

Activity 2: Data collection and analysis of consumer panel data from Nielsen datasets at the Kilts Center for Marketing

Document and disseminate the findings of processing the data from Nielsen datasets and extract the information needed (e.g., demographics, product market segment, etc.) for this project

Document and disseminate the findings of the data analysis obtained from Nielsen dataset to complement the data from social media

Activity 3: Collection and analysis of marketing cases and/or ads with unfairness and exclusions

Document and disseminate the findings of the analysis of unfair marketing cases and extract the features/forms of biases

Activity 4: Text mining and sentiment analysis of the collected data for the development of metrics of fairness in marketing, and the identification of customer-desired product features

Document the text mining and sentiment analysis for the data collected from social media

Establish and publish the method to identify the deciding factor/features that influence customers' judgement




Activity 1: Quantification and rating of bias and unfairness of marketing strategies and relate it to customer-desired product features

Document and disseminate the findings of researching the existing methods for fairness quantification and adverting parameterization

Define and document the quantification methods for the features identified from the data collected

Establish and publish the association/mapping between the marketing features and the product features

Activity 2: Network-based approach for choice modeling by incorporating customer preferences and perception to marketing (e.g., price) (un)fairness

Define and document the network-based model for choice prediction and demand forecasting

Incorporate customer-related attributes into the modeling

Establish and publish the modeling approach that can represent a marketing strategy into the customer-related attributes and product-related attributes

Activity 3: Validation of the network-based choice modeling via demand prediction

Validating the model with simulation studies on choice prediction

Validating the model with simulation studies on demand forecasting



Activity 1: Parameterize marketing strategies and incentive design for improved advertisement with fairness consideration

Document and disseminate the findings of the literature study on computational marketing and computational advertising

Create a computational design approach for marketing strategies with the inclusion of fairness parameters

Create different testing cases based on the design approach established


Activity 2: What-if scenario analysis with the simulation of recommended marketing strategies and quantify the impact of those strategies on brand value and demand

Finish the model testing with simulation studies

Develop the measures and metrics that can effectively evaluate the model performance and impacts of the devised marketing strategies

Finish what-if scenario analysis and summarize the results for conference presentation or journal publication

Activity 3: Validation through human-subject experiment and surveys

Create experimental protocols and the design of experiment; finish the pilot studies

Conducting online experiment and survey study via social media or Amazon Mechanical Turk

Finish the validation study on based on the experimental data and prepare the manuscript to publish the outcomes

Goal 3.6 (SA6) Privacy-Preserving Analytics in Health and Genomics Objective 3.6.a: Design and develop machine learning algorithms and software, and advanced security and privacy technologies, for privacy-preserving data analytics. Objective 3.6.b: Train, test and validate the models and algorithms with publicly available data and some controlled genomics and health data; develop innovative frameworks and practical privacy-preserving techniques. Objective 3.6.c: Test the algorithms and technologies to work with a wide range of data types and high-dimensional heterogeneous data sources; Develop and deploy bioinformatics workflows into the private cloud environment, the ARP.





Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Design and develop machine learning and deep learning algorithms and software for privacy-preserving data analytics; Data and Infrastructure request and preparation;

Document and disseminate the findings of literature research of privacy-preserving data analytics algorithms and software

Activity 2: Develop privacy-preserving analytics algorithms, which will be based on high-dimensional tensor mathematical optimization model and combinatorial models;

Initiate investigation on mathematical optimization models

Activity 3: The optimization models will be incorporated with machine learning and deep convolutional neural network models;

Document and disseminate findings on theoretical investigation of privacy preserving algorithms


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Train, test and validate the models and the algorithms with publicly available data and some controlled genomics and health data;

Complete the data collection of benchmark datasets and implementation of baselines


Activity 2: 2. Study and develop innovative frameworks and practical privacy-preserving techniques, to facilitate centralized as well as distributed data analysis;

Complete the development and implementation of the algorithms and frameworks


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: 1. Test the algorithms and technologies to work with a wide range of data types and high-dimensional heterogeneous data sources;

Complete the algorithms testing with high-dimensional heterogeneous data sources

Activity 2: Test these algorithms on available data sources from other research projects from this Track-I;

Complete the algorithms testing with real-world datasets

Activity 3: Deploy bioinformatics workflows into the private cloud environment, the Arkansas Research Platform ARP, proposed in this Track-I;

Complete the deployment of the software into the cloud environment


Goal 3.7 (SA7) Cryptography-Assisted Secure and Privacy-Preserving Learning Objective 3.7.a: Develop privacy-preserving federated learning methods by combining cryptography techniques and privacy models Objective 3.7.b: Explore how to protect the privacy of classification input data from the server hosting machine learning models Objective 3.7.c: Assess/Protect the trustworthiness of training data and machine learning models




Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: research the hybrid use of existing cryptography techniques and differential privacy in federated machine learning

A survey of existing cryptography techniques and their applications in differentially private federated learning

Activity 2: develop new applied cryptography techniques to use in combination with differential privacy for federated machine learning

Design of preliminary new cryptography techniques used for differentially private federated learning

Design of new cryptography techniques used for differentially private federated learning

Activity 3: develop unified security models for theoretical analysis of hybrid solutions

Preliminary united security models for analysis of hybrid solutions

United security models for analysis of hybrid solutions



Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: develop methods for building/perturbing the model so that it can respond to encrypted or perturbed classification input

Preliminary methods for building/perturbing model to support perturbed classification input

Methods for building/perturbing model to support perturbed classification input

Activity 2: study whether and how differential privacy can be achieved for classification input

Complete exploration of whether/how differential privacy can be achieved for classification input



Activity 1: develop methods to assess the trustworthiness of training data and machine learning models

Preliminary methods to assess the trustworthiness of training data and machine learning models

Methods to assess the trustworthiness of training data and machine learning models

Activity 2: develop methods to protect the trustworthiness of training data and machine learning models

Methods to protect the trustworthiness of training data and machine learning models


4. Social Media and Networks

4.1. Introduction

Social media and networking platforms have billions of active users and leverage significant impacts on society. New types of social media and networking platforms or new features of existing platforms continue to be developed to meet users’ demands. With an increasingly large amount of unstructured social data on these platforms, social media and networking analytics research has the following scientific challenges: 1) difficult to analytically assess collective impact of social media and networking on societal polarization and other social phenomenon due to fragmentation of debates and discussions in the existing social media and networking platforms; 2) lack of a central social media and networking platform for debate and discussions on important issues at national and international levels; 3) hard to detect mis/dis-information, how it disseminates, and assess its impact; 4) mining and using social media/networking data for logistics planning for disaster response; 5) content-based indexing of unstructured and multimedia data on social media platforms and towards an integration with decision-making systems through deep learning methods; and 6) arduous to visualize large social network data.


Below is the list of envisioned advancements to the state of the knowledge by each sub-theme (SM1 through SM4).

• SM1 will advance the state of the knowledge in argumentation polarization modeling by developing innovative quantitative opinion polarization techniques to model the formation and evolution of opinion polarization in large-scale cyber argumentation and deliberation with social networks.

• SM1 will elevate methods and techniques to predict individual or collective opinions on single or multiple solutions of issues using collaborative filtering and machine learning-based techniques.

• SM1 will improve social network methodology and social network sampling and data generation.

• The proposed research in SM2 advances our understanding of the role of Information and Communication Technology-mediated communications in the formation of emergent organizations with implications to business, marketing (explaining viral behaviors), and many other settings.

• The proposed research in SM2 is of particular interest to information scientists exploring the influence of social systems on user behaviors; studying ties between people, technology, and institutions; examining organizational structures, roles, and crowd processes; investigating the notions of individual and collective identities in a variety of information systems supporting crowdsourcing, citizen participation, eGovernance, crisis and disaster management, and several other manifestations of emergent organizations.

• The proposed research in SM2 would contribute to the theory of collective action to model the dynamics of deviant cyber behaviors, borrow from the literature on collective identity formation to explain the motivation needed to sustain such coordinated acts, assimilate factors pertaining to collective failures/success, and leverage notions of hypergraph to model complex (multidimensional and supra-dyadic) relations commonplace among members of deviant groups.

• The proposed research in SM2 advances the literature on cyber-collective actions and study the role of social media in organization and coordination of cyber social movements from individual, community, inter-organizational, and transnational perspectives.

• The proposed research in SM2 develops socio-computational predictive models that are efficient, reliable, scalable, explainable, reproducible, and theoretically grounded to help understand behaviors from social media platforms.


• Research performed in the sub-theme SM3 will allow decision making processes to utilize multi-source, heterogenous and multimodal data towards better performance.

• Research performed in the sub-theme SM3 will allow expanding the scope of learning and artificial intelligence techniques to multimedia data.

• Research performed in the sub-theme SM3 will afford applications such as disaster recovery to take advantage of vast, increasingly popular multimedia data that is often unstructured and insufficiently indexed.

• SM4 provides methods for indexing and fusing transportation infrastructure status data from a variety of sources (e.g., social media, satellite imagery, traffic camera videos).

• SM4 applies credibility detection methods for transportation infrastructure status data on social media.

• SM4 enables the use of near-real-time transportation infrastructure status data in logistics planning methods to support disaster response.

• SM4 creates new vehicle routing models and solution approaches for complex real-time logistics planning problems in disaster response.



SM1 Mining cyber argumentation data for collective opinions and their evolution

Zhan, Adams, S. Yang.

SM2 Socio-computational models for safer social media

Agarwal, Trudeau, Zhan, Milburn

SM3 Auto-annotation of multimedia data Dagtas, Trudeau, Milburn, Cothren

SM4 Informing disaster response with social media

Milburn, Dagtas, Liao, Zhan., Cothren., Talburt, Ussery, Nachtmann, Rainwater, Karim, Celebi

A detailed matrix of these goals, objectives, activities, and planned milestones over the five-year

period can be found in the Social Media and Networking Activity Matrix.


Goal 4.1 (SM1) Mining cyber argumentation data for collective opinions and their evolution Objective 4.1.a: Develop a cyber discourse social network platform Objective 4.1.b: Collect data using the developed cyber discourse social network platform Objective 4.1.c: Develop natural language processing algorithms to analyze discourse data collected by the platform and existing data

Goal 4.1 Output Metrics Publications, presentations, and reports (5): -- Five (5) peer-reviewed journal and/or conference papers (articles)

Assessments, questionnaires, and surveys (1): -- One (1) IRB-approved questionnaire for collecting discourse data

Software (1): -- One (1) cyber discourse social network platform

Datasets and algorithms (2): -- Two (2) advanced natural language algorithms (code with associated training and testing data)


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Brainstorming the features needed for the cyber discourse social network platform

Key features for the platform are determined

Activity 2: Software design for the platform

Software design document is finalized

Activity 3: Implement the platform

Platform is implemented

Activity 4: Test the platform The platform is tested Activity 4: Deploy and disseminate the platform

The platform is deployed and disseminated



Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Design the "hot button" questionnaire items

Develop questionnaire for collecting discourse data

Activity 2: Develop Individual and network question measures and submit IRB consent form

The question measures are determined, and IRB protocol is approved

Activity 3: Data collection by engaging students in the "hot button" issue discussions

The discussion data are collected, and two refereed articles are submitted/published


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Develop advanced natural language processing algorithms

The advanced natural language processing algorithms are devleoped

Activity 2: Test the natural language processing algorithms using the existing data

The algorithms are tested using existing data

Activity 3: Validate the natural language processing algorithms using the data collected by the developed platform

The algorithms are validated using the data collected from the platform and three referred articles are submitted/published


Goal 4.2 (SM2) Socio-computational models for safer social media Objective 4.2.a: Characterize online information environment (OIE) Objective 4.2.b: Develop socio-computational models to identify key actors and key groups of actors Objective 4.2.c: Study tactics, techniques, and procedures (TTPs) of deviant cyber campaigns Objective 4.2.d: Develop socio-computational models to measure power of a cyber campaign

Goal 4.2 Output Metrics Publications, presentations, and reports (4): -- One (1) taxonomy -- One (1) journal paper -- Two (2) conference presentations

Applications and platforms (1): -- Socio-computational models for OIE and TTP implemented in a web-based application

Datasets and algorithms (2): -- Two (2) socio-economic models and associated datasets



Activity 1: Study social media spaces and cyber campaigns to identify characteristics and features

-- Social media platforms identified -- Cyber campaigns identified -- Characteristics and features identified

Activity 2: Create a taxonomy of dimensions to characterize social media spaces

Taxonomy developed

Activity 3: Revisit and adjust taxonomy as social media space evolves

Revised taxonomy developed and published based on new social media, campaigns, features, and characteristics




Activity 1: Review cyber campaigns and social media data

-- Data sources identified -- Data acquisition procedures established Database setup

-- Data reviewed and modifications incorporated -- Data collected and shared with DART teams

-- Cyber campaign data reviewed, additional data collected if needed -- Data published according to NSF's data sharing policies and social media platforms' terms and agreements

Activity 2: Identify behavioral traits for key actors and key groups by leveraging OIE characterization

Key actors and key groups identified empirically

Behavioral traits of key actors and key groups identified

Activity 3: Develop computational model(s) for key actor and key group discovery

Model(s) developed

-- Model(s) refined Published in peer reviewed forums -- Model transitioned to usable web-based application

Activity 4: Evaluate model(s)

Model(s) evaluated, and refinements proposed



Activity 1: Review campaigns, social media platforms, and involved actors and groups

Campaigns identified for further review

Review of campaigns concluded with study methodology and findings published as case studies in peer reviewed forums


Activity 2: Identify and document tactics, techniques, and procedures (TTPs) (e.g., platform orchestration, botnets, inorganic behaviors, stalking, pacing, leading, threadjacking, hashtag latching, boosting, echo chambers)

TTPs identified and documented/published

Activity 3: Examine tactics, techniques, and procedures (TTPs) vis-a-vis OIE characterization

TTPs categorized based on the OIE taxonomy developed

TTP categorization revised based on behavior evolution



Activity 1: Review OIE characterization and TTPs to identify campaign attributes

Review of TTPs and OIE categorization completed

Campaign attributes identified to help assess its power

Activity 2: Develop computational model to measure power of a campaign by integrating attributes, key actors, key groups, collective action theory (and other theoretical constructs)

Model(s) developed

Model(s) refined Published in peer reviewed forums Model transitioned to usable web-based application

Activity 3: Evaluate model(s)

Model(s) evaluated, and refinements proposed


Goal 4.3 (SM3) Auto-annotation of multimedia data Objective 4.3.a: Develop multimedia indexing methods for social media data Objective 4.3.b: Design and implement deep learning methods for multimedia data Objective 4.3.c: Build Integrated smart applications based on unstructured multimedia data

Goal 4.3 Output Metrics Publications, Presentations, and Reports (3): -- Two (2) journal articles -- One (1) conference paper

Datasets and Algorithms (3): -- One (1) algorithm for Indexing -- One (1) algorithm for deep learning for multimedia data -- One (1) algorithm for integrated smart applications


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Define priorities and characteristics for multimedia data on social platforms

Key characteristics defined

Activity 2: Design and build algorithms for efficient retrieval of nontraditional data

Image, video, and 3D retrieval methods defined and tested

Activity 3: Advanced querying methods and implementation of interfaces with multimedia capabilities

Querying tools for multimedia data



Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Define learning objectives for social data from multimodal sources

Identify and define three major learning objectives document

Activity 2: Develop detection and classification methods

Object and event detection methods implemented

Activity 3: Deep learning applied to multimedia data and related indexing mechanisms

Two learning methods developed and tested

Activity 4: Verification of learning objectives and methods developed in Activity 2 and 3

Testing and reporting


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Define key applications for the implementation and testing of the indexing and retrieval mechanisms

Three key applications defined

Activity 2: Integration with disaster response and other applications defined in Activity 1

Integration methods developed

Activity 3: Define ethical and legal perspectives for the use of multimedia data

Use and access-based ethics principles defined for multimedia social data


Goal 4.4 (SM4) Informing disaster response with social media Objective 4.4.a: Extract and index content describing transportation infrastructure status from social platforms Objective 4.4.b: Fuse data from social platforms describing transportation infrastructure status with other data sources Objective 4.4.c: Assess credibility of data inputs from Objectives 4.4.a and 4.4.b Objective 4.4.d: Develop routing algorithms that use inputs from Objectives 4.4.a-4.4.c to support routing for disaster response

Goal 4.4 Output Metrics Publications, presentations, reports (8): -- Three (4) journal articles -- Two (2) conference papers -- Two (2) case studies of natural disasters scenarios

Applications and Platforms (1): -- GIS routing platform informed with social media feeds

Datasets and Algorithms (5): -- One (1) schema for mapping each datum to a probability describing its credibility -- Four (4) algorithms


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Study social platforms to identify types of content that describe transportation infrastructure status

Identify and define social platform content types of interest (e.g., image, video, text, etc.)

Activity 2: Develop and implement extraction techniques for identified types of social platform content

Develop social platform extraction techniques for content types of interest and pilot test on at least two disaster scenarios

Activity 3: Develop and implement indexing techniques for extracted social platform content

Develop and implement indexing techniques for extracted social platform content and pilot test on at least two disaster scenarios




Activity 1: Identify other data sources that contain real-time information regarding transportation infrastructure status

Identify and define content types of interest (e.g., satellite imagery, traffic cameras) from sources other than social platforms

Activity 2: Obtain and index transportation infrastructure data from other data sources

Obtain and index identified content types for at least two disaster scenarios

Activity 3: Develop and implement data fusion techniques to combine data from social platforms and other sources

Fuse data from social platforms and other sources for at least two disaster scenarios


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Develop and implement machine learning classifiers to detect quality of information

Obtain testing data from social platforms

Develop machine learning classifiers to detect false or low-quality information

Activity 2: Develop and implement schema to map credibility/quality scores for data to probabilistic inputs of transportation infrastructure status

Develop schema for mapping each datum to a probability describing its credibility




Activity 1: Identify critical routing problems with application in disaster response

Select at least two disaster response routing problem variants using Milburn's existing qualitative interview data

Activity 2: Develop models of identified disaster response routing problems and assess state of the literature

Conduct literature review for identified routing problem variants and publish journal article synthesizing review with qualitative data from 4.4.c.1

Activity 3: Develop and implement routing algorithms for identified routing problem variants

For at least two routing problem variants, develop, validate and test at least one solution algorithm each on randomly generated test networks

Publish two journal articles

Activity 4: Implement GIS testbed capable of displaying and analyzing real-time road status and routing algorithm outputs

Define GIS system requirements

Develop GIS system to display real-time road status inputs

Integrate routing algorithm outputs with GIS testbed

Integrate disaster scenario pilots with GIS testbed

Activity 5: Demonstrate models and solution approaches via pilot study of one or more disaster scenarios

Select one or more disaster scenarios for pilot

Obtain test data for selected disaster scenario

Test algorithm on real disaster scenario pilots

Publish journal article


5. Learning and Prediction Research in this topical area will focus on various techniques in prediction interpretation for large-

scale, deep learning using multi-source integrated data sets. Goal 5.1 Statistical Learning – Random Forests (RF) for Recurrent Event Analytics integrates the RF algorithm with classical statistical methods including the non-parametric MCF estimator and parametric NHPP, which will allow dynamic and unstructured covariate information to be incorporated into a tree-based method. Goal 5.2 Statistical Learning – Marked Temporal Point Process Enhancements via Long Short-Term Memory Networks combines the marked temporal point process with LSTM to model the dynamic event data without assuming any parametric forms. Goal 5.3 Deep Learning – Novel Approaches uses unsupervised learning to build a deep network from a series of shallow networks, each having simpler and more interpretable objective functions to cope with the stochastic nature real-life events, address the high dimensionality of data and action spaces, explore topological and group compression approaches, and investigate design and interpretability issues in deep reinforcement learning. Goal 5.4 Deep Learning – Efficiency and Specification develops new objective loss functions incorporated in any deep networks to solve million-scale problems to automatically and efficiently cluster easy- and hard- samples to optimize deep learning models and better model the distributions of deep features extracted from hard samples in more accurate ways. Goal 5.5 Harnessing Transaction Data through Feature Engineering investigates a framework of utilization transaction data for improved prediction and decision making in medicine and business with a particular goal to improve the identification of opioid use disorder and intervention timing by enhancing prediction and utilizing valuable information from transaction data.


A major challenge in building secure and widely adopted deep learning systems is that they sometimes make wrong, unexplainable, and/or unpredictable misclassifications. In addition to confusing examples of very different classes, they are also vulnerable to adversarial examples. These systems are often trained as large feed-forward error-backpropagating black boxes and thus we have no way of interpreting the meanings of their features and understanding the causes of misclassifications, a situation that can be exploited by attackers. Research in this theme will focus on applying statistical learning techniques alongside more advanced deep learning techniques to address three major challenges.

1. Violation of fundamental statistics principles: In big data environments where sample size is large (in some extreme cases, every single individual in a population can be observed), many basic principles in statistics need to be revised, such as the classical relationship between population and sample, the assumption of independent and identically distributed (iid) random variables, and so on. This challenge requires statistical learning methods in big data environments to be equipped with capabilities for addressing heterogeneity and hidden sub-populations within big datasets.

2. Mode specification and interpretation: With high-dimension, dynamic, unstructured covariate information which has been made possible by big data technologies, it is no longer realistic to specify parametric assumptions that adequately describe the complicated nonlinear dynamics between responses and covariate information. In addition, a certain amount of covariate information is almost always redundant in the statistical modeling or domain knowledge perspectives. Efficient variable selection using non-parametric methods has become one of the most sought-after capabilities in big data analytics.

3. Computing in big data environments: It is already a mainstream practice in today’s industry to store massive data on distributed big data platforms. This trend requires the statistical computing to be performed on distributed/parallelized computing nodes—an extremely critical issue but is often ignored in traditional statistical modeling.


We will investigate these challenges surrounding high-dimensional, dynamic, and unstructured data sets and explore solutions in the domains of genomics, transaction scenarios in eCommerce, and supply chain logistics.



LP1 Statistical Learning – Random Forests for Recurrent Event Analytics

Liu, X. Chimka, J.

LP2 Statistical Learning – Marked Temporal Point Process Enhancements via Long Short-Term Memory Networks

Rainwater, C. Liu, X.

LP3 Deep Learning – Novel Approaches Karim, Md

Celebi, E., Luu, K., Schrader, P., Kim, D., Kursun, O., Cheng, H., Alroobi, R.

LP4 Deep Learning – Efficiency and Specification Luu, K. Le, N., Rainwater, C.

LP5 Harnessing Transaction Data through Feature Engineering Zhang, S. Nachtmann, H.

The Learning and Prediction research theme will be implemented through the achievement of five major goals and their associated objectives and supporting activities. A detailed matrix of these goals, objectives, activities, and planned milestones over the five-year period can be found in the Learning and Prediction Activity Matrix.


Goal 5.1 (LP1) Statistical Learning – Random Forests for Recurrent Event Analytics Objective 5.1.a: Create the Random Forests for Recurrent Event Analytics, which integrates the RF algorithm with classical statistical methods allows dynamic feature information to be incorporated into a tree-based method. Objective 5.1.b: Create the Gradient Boosting method for Recurrent Event Analytics, which integrates the boost trees with classical statistical methods allows dynamic feature information. Objective 5.1.c: Perform comparison study between the methodologies above and identify future research directions

Goal 5.1 Output Metrics Publications, presentations, reports (7): -- Two (2) manuscripts submitted for publication -- Two (2) student theses or dissertations proposed -- One (1) student theses or dissertations defended -- One (1) submitted research proposals



Activity 1: Establish a preliminary model, and complete the theoretical investigation

Complete the preliminary theoretical investigation on the proposed modeling approach

Activity 2: Complete the coding and numerical examples; write, submit, revise paper

Complete the numerical studies, and submit a research paper

Activity 3: Revise paper and research outcomes dissemination through conferences

Complete the model improvement and paper revision




Activity 1: Establish a preliminary model, and complete the theoretical investigation under Obj 5,2

Complete the preliminary theoretical investigation on the proposed modeling approach

Activity 2: Complete the coding and numerical examples; write, submit, revise paper

Complete the numerical studies, and submit a research paper

Activity 3: Revise paper and disseminate research outcomes through conferences

Complete the model improvement and paper revision



Activity 1: Perform the comparison

Complete the numerical comparison study

Activity 2: Write, submit, revise paper

Submit a research paper

Activity 3: Identify future research directions

Identify the future research, and include the findings in research papers


Goal 5.2 (LP2) Statistical Learning – Marked Temporal Point Process (MTTP) Enhancements via LSTM Networks Objective 5.2.a: Develop methodology integrating the MTPP with LSTM Objective 5.2.b: Create scalable implementation of MTTP/LSTM approach applicable to real-world data analysis scenario Objective 5.2.c: Evaluate and assess MTTP/LSTM approach on real-world discrete data sets

Goal 5.2 Output Metrics Publications, presentations, reports (6): -- One (1) conference paper -- Two (2) conference presentations -- Two (2) journal publications -- One (1) case study

Workshops, demonstrations, and trainings (2): -- One (1) graduate seminar -- One (1) industry workshop


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Formally define approach integrating intensity function of MTTP into LTSM

Submit conference paper with initial model.

Activity 2: Establish proof-of-concept implementation of MTTP/LTSM approach

Present conference paper with preliminary results of implementation.

Submit journal article with conceptual findings and initial implementation of approach.

Activity 3: Perform benchmark of MTTP/LSTM tests on small simulated data sets

Publish white paper and GitHub repository with benchmark tests/results.



Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Assess data collected from Activities 1 and 2 of Obj 5.2c to define methodology computation performance requirements

Produce system requirements document for V2 implementation.

Activity 2: Create version 2 implementation of approach using lessons learned from Activity 2 of Obj 5.2a

Submitted conference paper

Present conference paper on V2 implementation progress.

Prepared manuscript of journal article on version 2 of large-scale implementation.

Activity 3: Validate performance of scalable implementation on large-scale simulated datasets

Present conference paper on V2 implementation progress.

Submitted journal article on version 2 of large-scale implementation.


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Acquire healthcare IoT datasets

Publish curated data to GitHub.

Activity 2: Acquire civil infrastructure datasets

Publish curated data to GitHub.

Activity 3: Establish baseline performance of predictions made by existing approaches applied to datasets from Obj 5.2c Activities 1 and 2

Publish white paper and GitHub repository with benchmark predictions.

Make conference presentation on baseline performance of predictions compared against existing approach.

Activity 4: Assess methodology implementation from Objective 2 on real-world datasets collected in Obj 5.2c Activities 1 and 2

Workshop with industry stakeholders providing data in Activities 1 and 2.

Prepared manuscript journal article covering practice-based computational efforts.


Goal 5.3 (LP3) Deep Learning – Novel Approaches Objective 5.3.a: Extract explanatory features from Deep Network Objective 5.3.b: Address high dimensionality issues in Deep Reinforcement Learning (DRL) using algebraic and topological methods Objective 5.3.c: Designing a novel rewarding model, and addressing interpretability issues in DRL

Goal 5.3 Output Metrics Publications, presentations, reports (10): -- Six (6) conference papers -- Two (2) Journal publication -- Two (2) Master's theses;

Workshops, demonstrations, and trainings (3): -- One (1) workshop -- One (1) teaching module -- One (1) Special topics undergraduate class offering

Datasets and algorithms (1): -- One (1) dataset



Activity 1: Development of novel self-supervised and flow-based deep learning approaches

Development of the first unsupervised convolutional area and the first flow-based deep learning approach.

Adaptation of the novel methods for application to the tactical agility dataset.

Development of the areal postprocessing of the self-supervised model and the Flow Autoencoder as deep graphical model with dual objectives.

Building of the deep stacked self-supervised model.

Adaptation of the developed novel deep networks to one of the datasets of the DC thrust.

Activity 2: Developing a library of classifiers for benchmarking.

Development of linear dimensionality reduction methods.

Development of the standard autoencoder method.

Development of signal feature extraction tools.

Development of the clustering/segmentation tools.

Development of the baseline classifier for the DC dataset.


Activity 3: Application of the developed methods on real-world datasets.

Application of the developed methods with applications on natural images and textures, and classification of a malware dataset.

Comparisons of the developed methods/libraries on the tactical agility dataset and malware dataset.

Exploratory studies with the datasets of the researchers in the DC thrust of DART. Evaluate various deep learning models on the malware dataset.

Demonstration of the quality of the features extracted by the self-supervised deep network and flow-based autoencoder methods on the tactical agility dataset.

Comparisons of the developed methods/ libraries on the dataset of the DC thrust. Develop the deployment platforms for the developed classification systems.



Activity 1: Investigate group theoretical, and topological properties of generalized neural network architectures

Investigate group theoretical approaches to generalized NN architecture design within the context of interpretability for Objectives 5.3a (Activity 1) and 5.3c (Activity 1)

Exploration of internal topologies in generalized NN structures using TDA and PH to identify architectures which address high dimensionality and enhance the developments in Objectives 5.3a (Activities 1 and 2) and Objective 5.3c (Activity 2).

Study of probabilistic spaces associated to generalized NN from the perspectives of group theory and topology for identifying structure that enhances inferential capabilities and address concerns in the activities of Objectives 5.3a and 5.3c (e.g., stochastic spiking).

Activity 2: Apply group and topological structures of neural networks in the context of DRL

Fusion of group theoretical and topological properties of generalized NN and their associated probability spaces for optimal robustness, data flexibility, and structural performance in the context of the activities of Objectives 5.3a and 5.3c.

Development of use cases of topological group structures in the context of DRL and its high dimensionality concerns involving Objectives 5.3a and 5.3c.




Activity 1: Design an improved reward process for DRL

Development of a generalized model of reward function in DRL addressing the issues with both sparse and dense feedback.

Development of use cases of the developed reward model.

Activity 2: Explore PH based filtering to optimize scenario space

Development of a PH based scenario-space optimization algorithm

Development of DRL models exploiting optimized scenario-space

Activity 3: Explore DRL interpretability

Causal inference and do-calculus modeling for DRL

Goal 5.4 (LP4) Deep Learning – Efficiency and Specification Objective 5.4.a: Create Novel Deep Learning Networks Executable with Reduced Computational Resources and Assess Performance Objective 5.4.b: Address Low-cost Deep Learning Algorithmic Analysis and Challenges Objective 5.4.c: Explore Low-cost Deep Learning Applications in Natural Images and Medical Images

Goal 5.4 Output Metrics Publications, presentations, reports (5): -- One (1) journal article -- Two (2) conference papers -- One (1) invited presentation at an Arkansas Institution -- One (1) invited presentation elsewhere

Workshops, demonstrations, and trainings (3): -- One (1) conference workshop -- One (1) tutorial -- One (1) teaching module




Activity 1: Develop and demonstrate new low-cost deep neural network algorithms.

-- Develop Teacher - Student Distillation Deep Learning Algorithms -- Develop Light-weight Deep Learning Algorithms

-- Develop Deep Network Compression Algorithms -- Develop Deep Network Pruning Algorithms

Activity 2: Develop new objective loss functions in deep neural networks

Develop auto and semi-auto deep network searching algorithms to discover optimal deep neural networks given a particular application and data training sets.

Deliver new novel deep network loss functions in combination with the deep networks to improve the performance

Activity 3: Develop low-cost deep learning methods for high-dimensional data.

Develop Depth-wise Separable Deep Network for Volumetric Data

-- Optimize and implement the deep networks on low-cost computers -- Fine-tuned and evaluated in on-the-edge devices.



Activity 1: Mathematically analyze the proposed deep learning methods

Develop analytic approaches to the proposed methods in Activities 1.1

Develop analytic approaches to the proposed methods in Activities 1.2

Activity 2: Improve the computational time and accuracy performance

Improve the computational time and accuracy performance on the standard databases and challenges

Improve the performance for the loss functions in deep learning


Activity 3: Analyze the complexity

Analyze the complexity of the presented low-cost deep learning methods when deploying on the dataset from the other thrusts in this project


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: The developed deep learning algorithms will be optimized and implemented in two applications, including natural images and medical imaging.

Develop Low-cost Deep Learning Approaches in Image Classification

Develop Low-cost Deep Learning Approaches in MRI Segmentation

Activity 2: The developed deep learning algorithms will be further optimized and deployed in high dimension data, such as: videos and medical MRI volumetric data

Develop Low-cost Deep Learning Approaches in Automatic Human Activity Recognition in videos.

Develop Memory-Effective Deep Network for Volumetric Data


Goal 5.5 (LP5) Harnessing Transaction Data through Feature Engineering Objective 5.5.a: Design advanced feature engineering techniques for high-dimensional temporal data Objective 5.5.b: Create an improved prediction and decision-making framework incorporating feature engineering with health transaction data Objective 5.5.c: Employ and validate the new framework for prediction and decision making with business transaction data

Goal 5.5 Output Metrics Publications, presentations, reports (7): -- Three (3) journal publications -- Three (3) conference presentations -- One (1) doctoral dissertation



Activity 1: Extract and process APCD data

Obtain and prepare cleaned data for research

Activity 2: Extract and engineer features from the high-dimensional temporal data

Acquire features that are highly representative

Activity 3: Explore and test automation of feature engineering in transaction data

Achieve automated feature engineering

Optimize feature engineering with transaction data

Improve robustness of automated feature engineering


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Develop deep learning prediction models and algorithms with feature engineering

Complete selection and testing of deep learning models

Improve the predictive models

Activity 2: Incorporate representation learning in prediction with engineered features

Implement and test autoencoders

Compare and test representation learning methods


Activity 3: Compare and validate prediction with existing models

Identify the best and robust predictive models


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Extract and process business transaction data

Obtain and prepare business transaction data

Activity 2: Employ feature engineering and prediction in the business transaction data

Employ and test various models for prediction

Establish the link between prediction and decision making

Activity 3: Validate the developed framework with the business transaction data

Finalize and promote a robust framework for transaction data


6. Education


Our vision is to create a model Data Science and Analytics program for colleges and universities in Arkansas to promote problem-based, and experiential-based pedagogy in critical thinking and analysis, technology familiarity, and foundation in math and statistics. This will form the basis of an educational ecosystem where learners receive a designed, consistent, sequenced, scaffolded, and modular education in data science with further educational and/or job opportunities available at appropriate points in their careers.

DART will foster the establishment of a Statewide Data Science Educational Ecosystem by: • Developing a combination of model programs, degrees, pedagogy, and curriculum including

a 9- week middle school coding block; a technical certificate, certificate of proficiency, and associate of science in data science; and a Bachelor of Science in data science with minors or concentrations.

• Providing resources and training for educators including $5,000 Seed Grants for project-related Education & Broadening Participation; Career Development Workshops for project participants and educators; and K12 teacher professional development on data science topics.

• Providing educational opportunities inside and outside the classroom for students. Undergraduate and graduate research assistantships in DART labs will be funded along with intensive data science and computing summer camps for undergraduates and research-based capstone projects and internships with industry partners.

• Ensuring broad participation to impact the pipeline of data science skilled workers through Summer Undergrad Research Experiences in DART labs for underserved students, scholarships for underserved students to the ASRI; and by connecting students to opportunities through the ACDS.


Programs in data science are not currently offered at most Arkansas IHEs. On the other hand, courses and programs in computer science and statistics are widely available across a spectrum of institutions. Through this project we expect to ensure that all collaborating IHEs will gain a better understanding of the nature of data science and the appropriate educational resources that such programs require. When the project is complete data science programs will be widespread across the state and will be available to all interested Arkansans.

6.3. Broader Societal Impacts:

Integrating data science research across the State and creating a deep and diverse data-ready workforce will pay immediate dividends in the form of increased federal grant funding, increased industrial research funding, and increased employment of well-paying jobs. As the State better aligns its investments with industry strengths and needs, more opportunities to improve the quality of life in Arkansas and steadily increase educational attainment and wages will develop. The HDR Big Idea recognizes that efforts in developing data cyberinfrastructure, education programs, and a deep workforce are most effective when linked to relevant data science research.

For the first time in AR EPSCoR history, we are working with three of the state’s HBCUs to ensure inclusive learning pathways for a broad population of students. The HBCU roles are critical to the success of the entire education component.

DART will include a data science summer institute for undergraduates, summer internships and research experiences, increased data science educational opportunities, and revamped curriculum to


include relevant data science topics and capstone projects. Developments in data cyberinfrastructure will increase sharing information among educational institutions, research institutions, and industry.



ED1

Contributing to the Data Science Educational Ecosystem by developing a combination of model programs, degrees, pedagogy, and curriculum including a 9-week middle school coding block; a technical certificate, certificate of proficiency, and associate of science in data science; and a Bachelor of Science in data science with minors or concentrations.

Shubert and Addison

Chowdhury, Scott, Siddique, Swiad, Fowler

The project team will create an informed, scalable, and replicable set of model programs, degrees,

pedagogy, and curriculum. This provides an excellent opportunity for the State’s colleges and universities to work together, identify key areas of concentration based on regional skills, needs, and interests, and optimize investments in people, facilities, and resources across academia, government, NGOs, and industry. The following activities and initiatives are designed to populate the front of the education pipeline with a rich, diverse set of students.

Middle School Coding Block DART will create a full 9-week curriculum for the middle school coding block in partnership with the

Arkansas School for Mathematics, Sciences, and the Arts (ASMSA), the Arkansas Computer Science Teacher Association, and the Arkansas Department of Education (ADE). A two-day workshop will be planned with K20 computer science educators to outline the curriculum and pedagogy and establish the project timeline, roles, and deliverables. The curriculum will contain built-in assessments and methodology for longitudinal tracking of outcomes and will be made publicly available and piloted in at least six schools by the end of Year 5.

Establishing Data Literacy and Responsibility Educating and developing a data-literate and data responsible workforce requires more than

modifying existing courses. Companies, government agencies, and NGOs need a workforce whose education, training, and experience includes domain of application, addressing ethical issues and social implications, and considering a hands-on practicum or capstone integrative experience. Pre-existing courses typically cover essential material but do so in the context of those majors necessitating new courses in an overall design that ensures program core curriculum continuity across the courses, internships, research, and years. Developing the base rigorous data science major, with concentrations, provides the basis from which associate degrees, minors, tracks, additional concentrations, technical certificates, etc., can be developed.

Courses developed for a rigorous data science major begin at an introductory level and progress through a series of new topics of increasing depth and complexity while remaining connected to previous courses, removing the scaffolding along the way. Partnering with institutions at all levels during the curriculum development process ensures smooth transitions for students from the 2- to 4-year colleges while simultaneously ensuring outcome and learning continuity. Specialization is likely to begin in the third and fourth years of a 4-year suggested plan of study which can be implemented through domain-focused concentrations. More advanced and/or domain specific programs (M.S., Ph.D.) can also be


developed leveraging the research programs and rigorous data science foundations at the undergraduate level. Designing, developing, and pilot implementing such a system program is proposed next.

Model Postsecondary Programs UAF recently developed a rigorous Bachelor of Science in Data Science with multiple concentrations

as its foundation. This program, cooperatively developed by the colleges of engineering, business, and arts and sciences, is designed with a core set of data science courses for all students (“hub”) and a set of concentrations for domain expertise (“spokes”) providing a means for adding concentrations (domains, “spokes”) as appropriate. The project team will use this program as the basis, reference point, and template for expanding postsecondary data science education throughout the State. In order to provide rigorous Data Science degrees, the focus on developing and expanding the program will be based on the employer workforce needs and a focus on creating a process and pipeline through the State’s higher education institutions that is rigorous, replicable, scalable, and achievable.

During the development of the UAF program, an employer needs survey was conducted with the help of an industry advisory committee. Employer input on workforce requirements and needs were strongly stated: the existing two-year, certificate, and master’s degree-level programs are not producing students who have the skills they need. While these graduates have the vocabulary and ability to use basic tools, they lack the rigorous foundations, training, and experience in critical thinking, analysis, synthesis, domain knowledge, and communication skills these companies need to be successful. And, there was no consistency across programs which compounded the lack of rigor. By designing an overall program at the state level, a key challenge in matching 2-year and 4-year curricula can be addressed. The program will align with recent National Academies of Sciences, Engineering, and Medicine (NASEM) recommendations to “evolve a range of educational pathways to prepare students for an array of data science roles in the workplace.”

A team of faculty from multiple colleges, departments, and campuses will use the UAF Data Science program as a base to develop a full range model postsecondary curriculum. The curriculum will consist of a set of core courses with options for electives that could vary by campus meeting the requirements for student outcomes, pedagogy, and core curriculum continuity. The pilot institutions include two- and four-year campuses around the state that will implement iterations of the larger model. Two-year campuses can choose to implement an associate of science degree or a technical certificate, while four-year institutions can choose any combination of technical certificate, associate degree, bachelor’s degree, or minor concentration. This method will ensure consistency in quality, availability, and transferability for students statewide while allowing institutions to maintain competitive advantages and autonomy.

UCA has domain specific concentrations in data science in Business, Computer Science, and Mathematics. Many of the aspects of the UAF B.S. program are already implemented in these distinct concentrations and faculty at UCA are currently working on packaging these courses into a standalone B.S. degree. Thus, early in the project there will be two four-year degree programs that will facilitate the development of two- and four-year programs across the state.

Integration with Research Projects in the Proposal A key opportunity in the design and development of the degree program will be to leverage the

topics in the research program as examples for the courses, directly and indirectly, and to integrate these into the curriculum, practicums, and senior design/capstone courses. Such topics include automated data curation, curating heterogenous data, antidiscrimination and preserving privacy in data.

Institutional Partnerships and Active Participation: 11 institutions have agreed to participate in implementing a technical certificate or associate of science in data science. Of the 11, two are HBCUs, and 10 are two-year campuses. Three other institutions, including an additional HBCU and two academic research institutions, will implement a B.S. in data science based on our model curriculum.


The collaborating institutions will be divided into three cohorts and the DART team will address institutional needs of each campus. The cohorts will then develop and implement their specific programs and apply for accreditation to the appropriate agencies.

Table 2: DART Participating Institutions; *Academic Research Institution; S Institution receiving subaward.

Institution Name Institution Type Region Project Component Arkansas State University, Beebe Two Year Public Central Education Arkansas State University, Mountain Home Two Year Public Northeast Education Arkansas State University S Public ARI* Northeast Research Black River Technical College Two Year Public Northeast Education Cossatot Community College of the U of A Two Year Public Southwest Education Northwest Arkansas Community College Two Year Public Northwest Education Philander Smith College (HBCU) S Four Year Private Central Education Shorter College (HBCU) S Two Year Private Central Education Southern Arkansas University Tech Two Year Public Southwest Education Southern Arkansas University S Public ARI* Southeast Education & Research U of A Community College at Batesville Two Year Public Northeast Education U of A Community College at Morrilton Two Year Public Central Education U of A Pulaski Technical College Two Year Public Central Education University of Arkansas for Medical Sciences S Public ARI* Central Research University of Arkansas, Fayetteville (MSI) S Public ARI* Northwest Education & Research University of Arkansas, Little Rock S Public ARI* Central Research University of Arkansas, Pine Bluff (HBCU) S Public ARI* Central Education & Research University of Central Arkansas S Public ARI* Central Education & Research


Goal 6 (ED 1) Developing a combination of model programs, degrees, pedagogy, and curriculum including a 9-week middle school coding block; a technical certificate, certificate of proficiency, and associate of science in data science; and a Bachelor of Science in data science with minors or concentrations.

Objective 6.1.a: Create a full 9-week curriculum for the middle school coding block to help struggling K12 teachers meet state coding requirement and provide rich training to K12 students Objective 6.1.b: Create a set of postsecondary programs of core courses with options for electives for a consistent set of Data Science Undergraduate degrees (e.g., Assoc. Degrees, 2+2 and "2, then 2"), Concentrations, and certificates


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Hold a two-day workshop to include K20 computer science educators to outline the curriculum and pedagogy and establish the project timeline, roles, and deliverables.

Workshop completed, plan finalized and

disseminated to stakeholders.

Activity 2: Develop curriculum

Curriculum 50%

complete Curriculum 100%

complete

Activity 3: Make curriculum publicly available through ADE and educational service co-ops

Curriculum posted on

multiple outlets

Activity 4: Pilot curriculum in at least 6 schools.

6 schools implement curriculum


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Create the 5-year Plan to meet the Objective

Plan disseminated to stakeholders

Review 5-yr plan & update as needed



Complete 5-yr plan

Activity 2: Identify the level of involvement and timing by academic institutions within the State

Cohorts identified; all collaborators assigned

Begin Cohort 1 Begin Cohort 2 &

Complete Cohort 1 Begin Cohort 3 &

Complete Cohort 2 Complete Cohort 3


Activity 3: Review UAF and UCA Data Science Programs with the Teams

1 meeting complete Update meeting complete

Update meeting complete

Activity 4: Convene workshops annually of engaged academic and government institutions to establish baseline

3 Workshops completed





Activity 5: Define Data Science Objectives and Outcomes base for defined degrees and certificates

Info disseminated to stakeholders

Updated Info

disseminated to stakeholders

Updated Info disseminated to stakeholders

Activity 6: Define Data Science Courses Objectives, Learning Outcomes, and applicability to the defined degrees and certificates


Updated Info


Activity 7: Dissemination of developed program details with collaborating institutions, government, and industry partners


Updated Info disseminated to

stakeholders

Updated Info


Activity 8: Ensure defined programs are in line with appropriate accrediting bodies

Identify "Wave 1" of accreditation candidates

Review "Wave 1" for

accreditation readiness

Identify "Wave 2" of accreditation candidates

Review "Wave 2" for accreditation readiness

Activity 9: Prepare and submit program proposals of each type at each level for appropriate approval

Begin "Cohort 1" Proposal Preparation

Submit "Cohort 1" Proposals


Submit "Cohort 2" Proposals


Activity 10: Evaluate progress and iteratively improve for future programs as appropriate

Evaluation report disseminated to

stakeholders

Evaluation report disseminated to stakeholders


Activity 11: When accreditation is available, propose first-pass consultative reviews by those bodies for ready institutions

"Wave 1" consultative

review

"Wave 2" consultative review

Activity 12: Prepare those institutions which do not have accredited programs in closely aligned areas for the pre-accreditation visit year

Non-accredited Cohort 1 pre-

accreditation review

Non-accredited Cohort 2 pre-accreditation review

Activity 13: Identify appropriate accreditation by program and provide visibility to academic institution administrations

Cohort 1 Appropriate Accreditation bodies

identified and reviewed with

appropriate academic institutions

Cohort 2 Appropriate Accreditation bodies

identified and reviewed with

appropriate academic institutions

Cohort 3 Appropriate Accreditation bodies identified and reviewed with appropriate academic institutions

Activity 14: Create and maintain clearing house for course materials

Create shared resources with UAF

UCA existing materials and

establish cataloging methodology

Add Cohort 1 developed materials

Update contributed materials



Activity 15: Connect students, courses, problems, data, etc., with the Research Themes

Identify "Opt-In" Research Theme Researchers &

Collaboration Types & Timing

Group 1 Collaboration Group 2 Collaboration Group 3 Collaboration Group 4 Collaboration


7. Workforce Development and Broadening Participation


The grand challenge of the workforce development and broadening participation initiatives is to create a larger, more diverse pipeline of people with rich educational experiences and skills in data science and computing graduating and entering the workforce in Arkansas. DART will provide a number of programs to address this challenge, including professional development opportunities for K20 educators, undergraduate and graduate students.



WD1

Provide K20 teacher and faculty opportunities for professional development spanning multiple disciplines

Fowler, Stanley -

WD2 Provide educational training opportunities inside and outside the classroom for students

Fowler, Stanley Schubert, Addison

WD3 Ensuring broad participation to impact the pipeline of data science skilled workers

Fowler, Stanley Schubert, Addison

Goal 7.1 Provide K20 teacher and faculty opportunities for professional development spanning

multiple disciplines. Objective 7.1.a. K-12 Teacher Professional Development: In partnership with the EAST Initiative,

DART will train K12 teachers throughout the state focusing on two areas: student leadership and emerging technology integration. Teachers will learn methods to integrate these new technologies into their classrooms, empower students to hone their leadership skills, and will receive technology for their classrooms. Each topical area will consist of two (2) one-day training sessions. In addition, a series of webinars will be developed to share best practices, disseminate updated resources, and provide support for troubleshooting technology. One teacher participant will be selected in years 2-5 to present about the project at the annual All-Hands meeting. The detailed implementation will be planned with EAST Initiative during Year 1.

Objective 7.1.b. Education & Broadening Participation Seed Grants: DART will solicit proposals for project related mini-seed grants for education, outreach, and broadening participation. Eligible entities will include school districts, post-secondary institutions, educational service co-ops, non-profits, or other entities supporting data science and computer science education and outreach activities in Arkansas. The central office will manage the solicitations which will be posted on the AEDC website, DART website, and emailed through higher education channels. The proposals will be reviewed by the central office, and outside experts as needed. One awardee will be selected each year to present their project at the annual All-Hands meeting.

Objective 7.1.c. Faculty Training: Provide funding for faculty at collaborating institutions to learn new skills and tools in data science so they can effectively teach the new data science curriculum developed in the education component. DART will identify faculty to be trained and assign them into cohorts. Training opportunities will be identified based on campus needs. One faculty member from each cohort will be invited to present about their experience at the annual All-Hands meeting.

Objective 7.1.d. Career Development Workshops: DART will hold regional and virtual workshops to enhance the research competitiveness of the state’s faculty in data science and computing. Workshop


topics will include science, grantsmanship, and commercialization. These will be tailored for students and early career faculty and will feature program officers from various agencies, entrepreneurs, and outside scientists. Three virtual workshops will be planned per year with possibility of face-to-face later in the project. Workshops will be free for attendees and open to the Arkansas higher education community. The workshops will also be recorded, and recordings will be made available to all DART participants.

Goal 7.2: Provide educational training opportunities inside and outside the classroom for students. Objective 7.2.a. Student Support at Participating Institutions: DART will support undergraduate

research assistants during the fall and spring semesters at each participating primarily undergraduate institution, and graduate research assistantships at the academic research institutions. This will provide students with the opportunity to work on real research problems provided by industry and will increase retention and broaden participation in the data science pipeline. All student participants will be invited to DART meetings as outlined in the communication plan below. In addition, DART will host student forums every other month to provide updates and receive feedback from the students regarding their participation and reinforce their success in the project. All student participants will also be invited to present their research in the annual Poster Competition, which will be judged by faculty, EAB, and IAB members. The winning students will receive travel awards to major national and regional conferences.

Objective 7.2.b: Summer Internships: DART will work with industry partners, including the industry advisory committee, to facilitate at least 5 internships annually during Years 2-5 for student participants at companies in relevant sectors and research centers. DART will work with the host companies and students to evaluate and iteratively improve the recruitment and hosting process. Dr. Addison and Dr. Schubert will lead this initiative. An application process will be developed with the industry partners and students will be recruited from all DART participating institutions.

Objective 7.2.c. Capstone Projects: Education team faculty will work with faculty from other project components to identify work that can be used as capstone-style projects for undergraduates at collaborating institutions. DART will identify at least 5 projects annually and publish at least three annually to the entire Arkansas higher education community.

Objective 7.2.d. Arkansas Summer Research Institute: The ASRI will be hosted in partnership with the Arkansas School for Mathematics, Sciences, and the Arts (ASMSA). The ASRI is an intensive professional development experience for STEM students (seniors in high school up to senior undergrads). This 1-week event will be offered in 2 sessions each year and will provide workshops, panel discussions, and research training activities. The main goals of ASRI are to a) build a diverse support network of peers for undergrads in Arkansas and b) provide professional development to STEM undergraduates. The ASRI will build on previous successful iterations but with a modified focus on Dart related topics. Each year’s program will be evaluated by the external evaluator. Participants will be added to ASRI alumni Facebook and LinkedIn groups to facilitate longitudinal tracking and communications.

Goal 7.3 Ensuring broad participation to impact the pipeline of data science skilled workers. Objective 7.3.a. Summer Undergraduate Research Experiences for underserved students: DART will

fund summer undergraduate research experiences (URE), for underserved students. Students will be recruited through established and successful campus-based programs like McNair and Arkansas Louis Stokes Alliance for Minority Participation (LSAMP), and by faculty at participating institutions. Faculty will apply for funds to host these students for 8 weeks, with a limit of $8,000 per award. Funds will support student stipends, housing, student-specific supplies, and in-state travel. Faculty are encouraged to host students who have completed the ASRI and to send recruited SURE students to the ASRI. The applications will be reviewed and awarded by the central office.

Objective 7.3.b. Scholarships to ASRI: DART will provide scholarships and targeted outreach to recruit URM students to attend the ASRI. Students will be recruited at all Arkansas higher education campuses, on social media, through email, and hopefully in-person recruiting events. DART will also


leverage relationships with LSAMP, the McNair Achievement Program, and other organizations that are connected with underserved students. Transportation stipends will be provided to these students as needed, as well as any technology or necessary supplies.

Objective 7.3.c. ACDS: DART will connect URM students to opportunities for internships, apprenticeships, jobs, and professional development through ACDS. Dr. Schubert and Dr. Addison will establish monthly meetings with Bill Yoder of ACDS where opportunities will be discussed. DART participants will be notified of opportunities in an email newsletter.


Goal 7.1 (WD) Provide K20 teacher and faculty opportunities for professional development spanning multiple disciplines. Objective 7.1.a: Enable K12 teachers to integrate new Computer Science/Data Science technologies into their classrooms, empower students to hone their leadership skills, and provide Objective 7.1.b: Provide opportunities to support education and broadening participation related activities in the fields of Computer Science/Information Science/Data Science at K20 levels and both formal and informal settings Objective 7.1.c: Provide funding for faculty at collaborating institutions to learn new skills and tools in data science so they can effectively teach the new curriculum. Objective 7.1.d: Host workshops to enhance the research competitiveness of the state’s faculty in data science and computing, particularly in grantsmanship and entrepreneurship technology for their classrooms.


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Host one training session annually, issue technology kits to teacher participants

1 Training session complete, kits issued




Activity 2: Host two support/training webinars annually

Two webinars completed




Activity 3: Establish platform for teachers to disseminate resources and troubleshoot

Platform implemented; past participants onboarded

New participants onboarded

New participants onboarded

Activity 4: Participate (booth or breakout) in EAST Initiative annual conference

1 Conference completed





Activity 5: Invite participating teachers to annual All-Hands

1 Teacher presentation at All Hands complete






Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Fund seed mini-grants annually at $5,000 each

10 seed grants awarded





Activity 2: Recipients attend Annual All-Hands

1 awardee presentation at All Hands complete






Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Identify faculty to be trained and assign to 5 cohorts

Cohort 1 established Cohort 2 established Cohort 3 established Cohort 4 established Cohort 5 established

Activity 2: Fund faculty annually at $5000 each for training

10 faculty trained 10 faculty trained 10 faculty trained 10 faculty trained 10 faculty trained


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Host annual workshops on a variety of grantsmanship and entrepreneurship topics







Goal 7.2 (WD) Provide educational training opportunities inside and outside the classroom for students. Objective 7.2.a: Student Support at Participating Institutions: Support undergraduate research assistants during the fall and spring semesters at each participating primarily undergraduate institution, and graduate research assistantships at the academic research institutions. Objective 7.2.b: Summer Internships: Facilitate industry internships for student participants at companies in relevant sectors and research centers. Objective 7.2.c: Connect with other research thrusts to develop relevant research-based capstone projects Objective 7.2.d: ASRI- intensive data science and computing summer camps for undergraduates


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Provide undergraduate research assistantships annually

15 UG supported 15 UG supported 15 UG supported 15 UG supported 15 UG supported

Activity 3: Provide graduate research assistantships annually

40 GA supported 40 GA supported 40 GA supported 40 GA supported 40 GA supported


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Identify internship opportunities for students at relevant companies

5 internships placed 5 internships placed 5 internships placed 5 internships placed

Activity 2: Follow up with hosting companies for feedback and evaluation

Develop intern and hosting company feedback and evaluation methodologies and instruments

Year 2 feedback and evaluation; iterative improvement for Year 3


Year 4 feedback and evaluation; iterative improvement for Year 5; Year 5 feedback and evaluation



Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Develop capstone projects annually

5+ capstones identified

5+ capstones identified 5+ capstones identified

5+ capstones identified 5+ capstones identified

Activity 2: Disseminate projects to all collaborating institutions

3+ capstones published





Activity 3: Invite capstone students to present at annual All-Hands

1 student presentation at All-Hands complete





Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Host ASRI Annually and invite all DART undergrads

1 ASRI Complete 1 ASRI Complete 1 ASRI Complete 1 ASRI Complete 1 ASRI Complete

Activity 2: Evaluate and revise programming based on student and presenter feedback


stakeholders


stakeholders


stakeholders

Evaluation report disseminated to stakeholders

Goal 7.3 (WD) Ensuring broad participation to impact the pipeline of data science skilled workers Objective 7.3.a: Summer Undergraduate Research Experiences for underserved students: Fund summer undergraduate research experiences (URE), for underserved students Objective 7.3.b: Scholarships for underserved students to the ASRI Objective 7.3.c: Connecting students to opportunities through the ACDS


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Provide summer UREs to URM students annually

10 UG supported 10 UG supported 10 UG supported 10 UG supported 10 UG supported


Activity 2: Students participate in annual All-Hands meeting poster competition

1 poster competition complete





Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Provide scholarships/recruit students annually

20+ scholarships provided






Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Co-host statewide workshops on Data Science topics

1+ workshop completed





Activity 2: Collaborate on Data Science apprenticeship programs- recruiting partners and developing curriculum

Develop apprentice and hosting company feedback and evaluation methodologies and instruments




Year 4 feedback and evaluation; iterative improvement for Year 5; Year 5 feedback and evaluation


8. Communication and Dissemination

8.1. Introduction

The grand challenge of the communication and dissemination effort is to ensure that all DART participants are aware of their role and responsibilities to the project, and to make the public aware of DART and its success.

8.2. Implementation Plan


CD1 Maintain interproject communication to accomplish milestones and relay updates Fowler, Stanley Ford, Cothren

CD2 Educate the public about DART accomplishments

Fowler, Stanley Ford, Cothren

Goal 8.1: Maintain interproject communication to accomplish milestones and relay updates. Objective 8.1.a: Day to Day Communication: Daily project-related communication will take place

mostly via email and GitLab. If during the first year the project faces challenges with these two platforms, other platforms like Slack will be explored. DART will also hold virtual office hours monthly or as needed where participants can drop in and ask any project related questions. The office hours will be staffed by representatives from the SSC and central office.

Objective 8.1.b.- Monthly Webinars & Component Meetings: DART will hold monthly webinars beginning in September 2020. Each webinar will have a short presentation followed by open discussion and questions. DART monthly webinars will be on the 3rd Wednesday of each month, in the same time period as the SSC monthly meeting. The topics will rotate between project management items like reporting and overviews of DART as a project, as well as specific research topics, important research results, and more. Each research component will also meet as a team monthly to communicate needs and progress. These meetings will mainly occur online via Zoom or Microsoft Teams and similar platforms.

Objective 8.1.c- Face-to-Face Meetings: Two project-wide face-to-face meetings per year will be hosted. The Annual All-hands Meeting and Poster Competition will be attended by all project faculty, students, industry partners, administrative committee members, evaluators, and external advisory board members. The Annual Retreat will be for faculty and graduate student participants. These meetings will facilitate team building and foster a sense of collaboration among the group. The central office is responsible for the logistical planning of these two events. The retreat will take place in a central location and the All-Hands meeting location will be rotated among the regions of the state. Online participation will be facilitated for both meetings for participants who cannot join in person.

Goal 8.2: Educate the public about DART accomplishments. Objective 8.2.a.- Maintain public-facing communication outlets to inform public about DART: DART

will establish a public project website that contains general information about the project, important research results, contact people, a list of all participants, links to faculty websites, and other content that will be refreshed at least quarterly. The central office will also maintain the existing @arepscor Facebook, Twitter, and YouTube pages with DART related content. Blogs will be posted with important updates related to DART quarterly on the AEDC website.

Objective 8.2.b.- Campus Communications: A project-wide communications team composed of communications staff from each participating institution, including AEDC, will be created. The communications team will use uniform citations and branding for all project-related releases. AEDC will issue press releases and blog posts related to overall project success, special events, and seed grant opportunities. The communications team will work together to release other pertinent information like new grant awards, patents, publications, and other highlights from each campus.


Objective 8.2.c.- Project Data: Project data will be submitted by participants to the project’s internal reporting system, ER Core. Mandatory NSF reporting data will be collected, as well as additional information like startup companies and other major accomplishments. Participants will be encouraged to enter data throughout the year into ER Core, and there will be a cutoff date before annual reporting each year for data submission for the report. The central office will establish and maintain this site.

Objective 8.2.d.- Technical dissemination channels: Project faculty will submit journal articles to scientific publications associated with data science and computing. Funds for travel stipends will be reserved to send students and faculty to national meetings related to data science and computer science research and education. Impacts and significant findings of research activities will be presented at these meetings. Relevant meetings include national meetings for professional societies and industry meetings. All outputs related to this will be reported in ER Core. See individual research component matrices for specific dissemination output goals.

Objective 8.2.e.- Science Journalism Challenge: Beginning in year 2, the project will host a statewide science reporting challenge. Journalism students (undergraduate and graduate level) from Arkansas institutions may apply to win first, second, or third place awards. Applicants will be paired with a project participant student and faculty member to develop a story about relevant research. Submissions will be evaluated by a committee of Arkansas public relations professionals, including representatives from major advertising agencies and news outlets. The first-place award will include publication in a major Arkansas news outlet. Award recipients will also be invited to present their stories at the annual all-hands meeting.


Goal 8.1 Maintain interproject communication to accomplish milestones and relay updates Objective 8.1.a: Day to Day Communication: Daily project-related communication will take place mostly via email and GitLab. If during the first year the project faces challenges with these two platforms, other platforms like Slack will be explored Objective 8.1.b: Monthly Webinars & Component Meetings: Monthly webinar meetings will be held to share updates, events, and other news between faculty, students, administrators, and industry partners. Components will also meet monthly to manage project milestones and activities. Objective 8.1.c: Face-to-Face Meetings: Two project-wide face-to-face meetings per year will be hosted. The Annual All-hands Meeting and Poster Competition will be attended by all project faculty, students, industry partners, administrative committee members, evaluators, and external advisory board members. The Annual Retreat will be for faculty and graduate student participants. These meetings will facilitate team building and foster a sense of collaboration among the group.


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Establish platform to maintain daily communication

Platform established and participants onboarded


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Host DART topical webinars monthly

11 webinars complete 11 webinars complete 11 webinars complete 11 webinars complete 11 webinars complete

Activity 2: Host monthly component team meetings

11 meetings per component (6 components)






Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Host annual All-Hands meeting & poster competition

1 All Hands & Poster Competition complete





Activity 2: Host annual retreat for faculty and grad students

1 Retreat completed 1 Retreat completed 1 Retreat completed 1 Retreat completed 1 Retreat completed


Goal 8.2 Educate the public about DART accomplishments Objective 8.2.a: Maintain public-facing communication outlets to inform public about DART Objective 8.2.b: Campus Communications: A project-wide communications team composed of communications staff from each participating institution, including AEDC, will be created. The communications team will use uniform citations and branding for all project-related releases. AEDC will issue press releases and blog posts related to overall project success, special events, and seed grant opportunities. The communications team will work together to release other pertinent information like new grant awards, patents, publications, and other highlights from each campus. Objective 8.2.c: Project Data: Project data will be submitted by participants to the project’s internal reporting system, ER Core. Mandatory NSF reporting data will be collected, as well as additional information like startup companies and other major accomplishments. Objective 8.2.d: Technical dissemination channels: Project faculty will submit journal articles to scientific publications associated with data science and computing. Funds for travel stipends will be reserved to send students and faculty to national meetings related to data science and computer science research and education. Impacts and significant findings of research activities will be presented at these meetings. Relevant meetings include national meetings for professional societies and industry meetings. Objective 8.2.e: Science Journalism Challenge: Beginning in year 2, the project will host a statewide science reporting challenge. Journalism students (undergraduate and graduate level) from Arkansas institutions may apply to win first, second, or third place awards. Applicants will be paired with a project participant student and faculty member to develop a story about relevant research. Submissions will be evaluated by a committee of Arkansas public relations professionals, including representatives from major advertising agencies and news outlets. The first-place award will include publication in a major Arkansas news outlet. Award recipients will also be invited to present their stories at the annual all-hands meeting.


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Establish project website

Project website published

Activity 2: Maintain project website and refresh content at least quarterly

Content posted Content posted Content posted Content posted

Activity 3: Publish quarterly blog posts about DART on AEDC blog

4 blogs published 4 blogs published 4 blogs published 4 blogs published 4 blogs published


Activity 4: Maintain @arepscor Facebook, Twitter, and YouTube channels and refresh DART content frequently

Following increased by 10%






Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Establish listserv and group of communications reps from each participating campus

Committee formed; first meeting complete

Activity 2: Hold annual check-in meetings to ensure proper citation of project and related messaging and disseminate project updates

1 meeting complete 1 meeting complete 1 meeting complete 1 meeting complete


Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Establish DART ER Core Site

ER Core Site published & accessible

Activity 2: Maintain DART ER Core site and provide annual training to participants

Participants onboarded; 3 training webinars complete

3 training webinars complete





Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Presenting at national conferences / professional societies

2 presentations complete




Activity 2: Publications 1 publication complete 2 publications complete

3 publications complete



Activity 3: Statewide Workshops for Cohorts and Waves

2 Workshops complete 2 Workshops complete 2 Workshops complete 2 Workshops complete 2 Workshops complete

Objective 8.2.e Specific Milestones

Year 1 Year 2 Year 3 Year 4 Year 5 Activity 1: Create committee responsible for implementing SJC

Committee formed, first meeting complete

Activity 2: Develop plan for SJC

Plan disseminated to stakeholders

Activity 3: Host annual SJC 1 competition completed

1 competition completed



Activity 4: Invite SJC winners to present at annual All-Hands

1 presentation at All Hands completed





9. Appendix A: Project SWOT Table

Likelihood of Occurrence

Impact Challenge/Risk Mitigation Steps

High High Faculty overwhelmed by new course delivery and less time for research

Reduced meeting and internal reporting requirements in year 1 to allow focus on research; AEDC follow-up with schools to ensure adequate release time

High High Enabling technologies are advancing rapidly so teams will need to stay ahead of current trends

Stay abreast of industry advancements (and keep an eye on academic work)

High High Lack of human infrastructure to teach Data Science courses at collaborating campuses

Use the 'Oxford Model' of tutorials - one-on-one instructors, remotely working with students

High Medium Personnel turnover, both faculty and students

Avoid single points of failure, always have at least two people up-to-speed on each critical task. Need good reports and documentation of everyone’s work and progress. Work on keeping people happy, and feeling engaged. Make all feel like they are part of something larger.

High Medium Matching human perception with machine learning and auto-annotation techniques remain a huge challenge (e.g. face recognition vs. mood recognition).

Use a reduced data set along with proper data compression and management techniques

Medium High Difficulty in recruiting international, and possibly full-time domestic, GRA students

Work with graduate school committees to ensure appropriate outcomes

Medium High Hiring freezes Work with school administrations to give priority openings supporting DART

Medium High Lack of funded management roles in DART Work with campuses to identify persons will to help in-kind as management resources

Medium High Lack of diversity Strive to hire more diversity when openings need to be filled

Medium High Some research teams have not worked together before which may lead to team management challenges

Regular meetings, joint team assignments, collaborations




Medium High Rapidly changing and very competitive subfields of privacy, security, and fairness in AI

Stay on and ahead of current trends and adapt research activities. Attend virtual conferences early, followed by in-person attendance where possible. Collaborate with researchers outside the state.

Low High

SM4 requires broad collaboration and use of outputs from SM3, Research Theme DC, and Research Theme LP in order to be successful. Specifically, the novel vehicle routing models and solution approaches require novel inputs from the various big data methodologies outlined in the overall project plan.

SM4 lead will proactively communicate with PIs from SM3, DC and LP beginning in Year 1 and throughout the project duration. Establish liaison relationships across these Goals.

Medium Medium

Lack of federated identity across participating institutions. Possible with UA system but not others. Creates friction trying to data/code resources across Globus and GitLab

Regular meetings with the 'right people' across the Institutes to make sure all know what's going on.

Medium Medium

Tradeoff of conducting cutting edge research on social awareness and implementing known algorithms, software, platform to address social concerns

Discuss with project team to balance intellectual merit and broader impacts

Medium Medium In SM, minimizing impacts of fake and misleading information dissemination is challenging.

To reduce the impact, a hybrid model of assessing collective argument credibility will be developed.

Medium Medium Improving the quality of discourse is challenging.

Mining unstructured data from our cyber argumentation and social networking platform to analyze its various characterizations, such as user participation, community structure, user’s influence, people-pleasers or devil’s advocates, and diversity aware social connection recommendation.

Medium Medium Social media data collection is often challenged by API rate limits. This may impact our current data collection methodology.

Data acquisition cost has been budgeted in the SM research theme to obtain social media data. Also, modifying the frequency of data collection and using a parallel/distributed data collection architecture will help mitigate any API rate limit related issues.




Medium Medium Maintaining privacy of the online identities collected from social media data.

Before conducting the research, PI will submit the study protocol to university's Institutional Review Board (IRB) for ethics and safety approval. PI has received approval from his university's IRB for other studies involving similar data collection methodologies from social media platforms. PI has completed the CITI program courses on Social and Behavioral Responsible Conduct of Research, Human Research, and Social and Behavioral Research Investigators and Key Personnel that expire on May 8, 2023.

Medium Medium Indexing and annotation mechanisms may not match the level and nature of those required by advanced learning techniques.

Multi-modal annotation mechanisms that combine existing text-based data will augment the techniques developed to improve the accuracy and performance.

Medium Medium

Obtaining transportation infrastructure data from a variety of data sources (e.g., social media, satellite imagery, traffic camera videos) for the same disaster event, so that the data can be fused to predict infrastructure status.

Purchase of social media data is part of the project plan and budget for Research Theme SM; will pursue the procurement of other data types from transportation and emergency management officials.

Medium Low

Considering that the project is ambitious (it strives for a high level of quality, verifiability, science and productivity for a short period of realization) while depending on many external factors, it is possible to deviate from the schedule.

Elaboration of a risk assessment for the delay in the schedule, followed by plans for their suppression. Establishing a project implementation monitoring system on a weekly basis. Risk assessment will continue during project’s course and mitigation strategies will be developed.

Medium Low Computation and storage requirements of the techniques and data sets may go beyond the capabilities of the systems and equipment used.

Learning and indexing mechanisms will initially use a hybrid approach to adapt algorithms to user expectations and ground truth gradually.


10. Appendix B: Project and Theme-specific Logic Models Project-wide, long-term Outcomes (Impacts):

-- Research contributions will improve the learning and prediction of data in a spectrum of applications including commerce, cybersecurity, disaster and emergency management, energy, environment, healthcare, retail, and transportation

-- Research outputs will generate interest in data science and help engage, encourage, and recruit a broad spectrum of learners as well as researchers

-- Arkansas should see a growth in research and education initiatives in data science creating a large number of diverse pipelines for data scientists, engineers, and technicians

-- DART will grow the segment of society that can benefit from Artificial Intelligence-driven solutions by eliminating economic barriers to technology access and boost Artificial Intelligence applications and efficient platforms to support Arkansas economy and workforce development


Research Theme

Input(s) Objectives Output(s) Short-term Outcomes Medium Term Outcomes

Research Theme 1: Cyber Infrastructure

Goal 1 (CI1) Staff: Cothren, Prior, Springer, Chaffin, Tarbox, Deaton, DuRousseau, Pummill, Merrifield Partnerships: Great Plains Network

Objective 1.1.a: Establish the Arkansas Research Computing Collaborative (ARCC)

Hardware and Software Infrastructure: -- Install, configure, and make available data science nodes on Pinnacle Portal -- ScienceDMZ at UA and UAMS/UALR -- 100Gb connection between ScienceDMZs. -- Establish dedicated DART GitLab repository -- Setup Globus data management services to point at DART storage arrays Documentation and User Guides: -- Create a technical management document defining organizational

-- 100% increase in active accounts on Pinnacle and Grace -- 100% increase in overall use measured in gigflops -- Code developed by DART researchers is shared via GitLab repository linked to public GitHub -- Sharing of large data sets among individuals, institutions, and HPC clusters.

-- enhanced academic collaboration across Arkansas campuses measured by increased publications using CI resources -- enhanced collaboration between industry and academia measured by the number of such projects that use CI resources -- trusted data sharing between collaborating partners measured by the number of recorded data transfers and the total number of TB transferred

Objective 1.1.b: Upgrade cluster for data science research activity and integrate with existing resources


Research Theme


Objective 1.1.c: Establish a Little Rock (UAMS, UALR) ScienceDMZ and high-speed connection with UAMS

structure, roles, and responsibilities of ARCC for personnel at participating campuses -- Amend existing MOU for ARCC expansion -- UAF and UAMS will create CI Plans to support DART (1 x UAF, 1 x UAMS); these CI Plans will serve as templates for other Institutions -- Create and publish document outlining GitLab user guidelines and minimum standard for code repository Workshops, demonstrations, and trainings: -- Two (2) online workshops per year for onboarding to ARP resources in YR2-5 -- Five (5) online software carpentry workshops per year in YR2-5 focusing on developing and sharing code and data; data

Objective 1.1.d: Establish a data and code sharing environment (GitLab and Globus)


Research Theme


Objective 1.1.e: Establish necessary controls to store and manage controlled unclassified, HIPAA-related, and proprietary information at UA and UAMS (other institutions if possible)

science programming; and data management -- Train and certify two (2) new software carpentry instructors (across the jurisdiction) per year in YR2-5 Applications and platforms: -- Create one (1) distributed computing testbed for HDFS, Apache Spark, others (DC) - -Create four (4) spatiotemporal testbeds for (CI/DC/SM/LP)

Goal 2 (CI2) Staff: Springer, Conde, Huff, Milanova Equipment: As determined by need and existing capability

Objective 1.2.a: Investigate state-of-the-art visualization solutions

Workshops, demonstrations, and trainings: -- One (1) online workshops per year for advanced visualization in YR2-5


Research Theme


Objective 1.2.b: Define domain-specific integration of visualization solutions

Publications, presentations, and reports: -- Three (3) presentations, reports, or other publications: 1 in YR1 and 2 in YR 2 Applications and platforms: -- Develop one (1) visualization solution for each research theme, including CI (5 total) -- Integrate one (1) visualization into existing testbed for each research theme (4 total)

Objective 1.2.c: Introduce/integrate visualization for shared test beds

Research Theme 2: Data Life Cycle and Curation

Goal 2.1 (DC1) Staff: Talburt, Cothren, Liao, Liu, Rainwater, Tudoreanu, Ussery, Wang, Xu, Yang Partnerships: UAMS, UAF. UALR Facilities: Campus offices, labs, IT Services and Networking

Objective 2.1.a: Automate Reference Clustering / Automate Data Quality Assessment

Workshops, demonstrations, and trainings: -- Five (5) conference workshops Publications, presentations, and reports: -- 14 research publications describing

-- Citations of papers and reports -- DART Git repository accesses, downloads, and branches show acceptance and use of algorithms within DART -- Public Git repository accesses, downloads, and branches show adoption of algorithms in the data

-- Use of HDFS as a platform for data curation increases in industry -- The balance between manual and automated data curation, as evidenced by industry and academic articles, tilts toward unsupervised automation in the data

Objective 2.1.b: Automate Data Cleansing


Research Theme


Equipment: Local Resources Infrastructure: Globus File Sharing, GitHub

Objective 2.1.c: Automate Data Integration

new methods and processes -- 11 journal and conference publications -- 16 presentation -- 5 PhD dissertations -- 2 potential patents and business incubation Datasets and algorithms: -- Nine (9) Mathematical formulations and algorithms for optimal collaborative data collection and cleansing. -- Seven (7) Mathematical formulations, algorithms and statistical methods for need- and prediction-based data collection, and scalable decision making

science community -- Data science programs in the state adopt some new methodologies and algorithms in coursework

curation process -- Genome and proteome quality scores become commonly accepted standards as evidenced in common repositories of that data. -- Good practice, as evidenced in government and industry standards, prohibit sending data without encapsulating a complete explanation of it syntax, semantics, and data quality requirements

Goal 2.2 (DC2) Staff: Talburt, Wang, Tudoreanu, Pierce, Liu, Rainwater Partnerships: UAMS, UAF. UALR Facilities: Campus offices, labs, IT Services and Networking Equipment: Local

Objective 2.2: Build a POC for Positive Data Control (PDC)

Workshops, demonstrations, and trainings: -- Three (3) demonstrations of data exchange between PDC Systems

-- Citations of papers and reports -- DART Git repository accesses, downloads, and branches show acceptance and use of algorithms within DART -- Public Git repository accesses, downloads, and branches show adoption of algorithms in the data


Research Theme


Resources Infrastructure: Globus File Sharing, GitHub

Publications, presentations, and reports: -- Three (3) conference presentations -- Three (3) research papers Applications and platforms: -- One (1) working prototype of a positive data control system -- One (1) novel design of a system to control data access and movement in synchronization with a data catalog and data governance standards

science community -- Read/write firewalls around enterprise data mirror "external connection firewalls" around the enterprise intranet.

Goal 2.3 (DC3) Staff: Ussery, Byrum, Jun, Yang, Liu, Rainwater Partnerships: UAMS, UAF, UALR Facilities: UAMS/UA HPC infrastructure Equipment: EMC object store on the UAMS HPC

Objective 2.3.a: Standardize pipelines for genome and proteome storage, retrieval, and visualization

Workshops, demonstrations, and trainings: -- Host one (1) workshop on how to use these standardized pipelines Publications, presentations, and reports:

-- Citations of papers and reports -- DART Git repository accesses, downloads, and branches show acceptance and use of algorithms within DART -- Public Git repository accesses, downloads, and branches show adoption


Research Theme


Infrastructure: Globus File Sharing, GitHub

Objective 2.3.b: Automate quality scores for biological sequence data

-- Five (5) papers published on standardized pipelines for genomics and proteomics. -- Five (5) conference presentations -- Five (5) Journal publications -- Two (2) PhD dissertations Datasets and algorithms: -- At least one (1) standardized database with genomics and proteomics quality scores shared with DART researchers -- One (1) algorithm (code with associated training and testing data)

of algorithms in the data science community-- Pipelines for storage and retrieval of millions of genomes and proteomes are possible. -- Visualizing massive amounts of biological sequence data is possible -- New data quality dimensions and metrics are in use. Objective 2.3.c: Apply

machine learning methods to systems biology

Research Theme 3: Social Awareness

Goal 3.1 (SA1) Staff: Xintao Wu, Qinghua Li, Anna Zajiciek

Objective 3.1.a: Identify potential vulnerabilities of deep learning algorithms

Workshops, demonstrations, and trainings: -- One (1) tutorial given at major AI conference Publications, presentations, and reports: -- Two (2) conference papers -- One (1) journal paper

-- Citations of papers and reports -- DART Git repository accesses, downloads, and branches show acceptance and use of algorithms within DART -- Public Git repository accesses, downloads, and branches show adoption of algorithms in the data science community

-- Public debate and policy, as evidenced in press reporting, government regulations, and company policy discusses the methods, algorithms, and findings related to social awareness -- Channels for collecting and publishing personal

Objective 3.1.b: Develop a universal threat- and privacy-aware deep learning framework


Research Theme


Objective 3.1.c: Conduct comprehensive evaluations of the proposed framework and models

-- One (1) thesis -- One (1) proposal

information and sources of breaches are reported and adopted by the general public -- Privacy preserving data analytics in genomics and health data analytics are evident. -- Influence policies for developing and adopting cryptography based privacy protection models and change the practice of insufficient trust evaluation/enforcement for machine learning

Goal 3.2 (SA2) Staff: Chenyi Hu, Ningning Wu, Xintao Wu

Objective 3.2.a: Improve crowdsourcing data quality with considerations of uncertainty

Publications, presentations, and reports: -- Two (2) conference papers -- One (1) journal paper -- One (1) thesis -- One (1) proposal

Objective 3.2.b: Enhance available inference and learning models with novel algorithms for improved effectiveness and efficiency Objective 3.2.c: Verify and validate the robustness and trustworthiness of information from crowdsourcing data

Goal 3.3 (SA3) Staff: Ningning Wu, Qinhua Li, Chenyi Hu, Xintao Wu

Objective 3.3.a: Investigate on personal identifying information and their privacy issues


Objective 3.3.b: Investigate appropriate multimodal deep learning techniques to identify discriminative and stigmatizing information


Research Theme


Objective 3.3.c: Develop a user-centric privacy monitoring and protection framework

Goal 3.4 (SA4) Staff: Lu Zhang, Xintao Wu, Zhenghui Sha, Anna Zajicek

Objective 3.4.a: Explore deep learning-based techniques to detect cross-media discrimination.

Workshops, demonstrations, and trainings: -- One (1) tutorial given at major AI conference Publications, presentations, and reports: -- Two (2) conference papers -- One (1) journal paper -- One (1) thesis -- One (1) proposal Datasets and algorithms: -- One (1) algorithm (code with associated training and testing data)

Objective 3.4.b: Design generative adversarial models to remove cross-media discrimination. Objective 3.4.c: Develop a joint multi-modal deep learning framework to detect and prevent cross-media discrimination. Test and evaluate the proposed techniques and models with large-scale social media data.


Research Theme


Goal 3.5 (SA5) Staff: Zhenghui Sha, Lu Zhang, Xintao Wu

Objective 3.5.a: Text mining and sentiment analysis of user-generated data from social media and consumer shopping records to extract customer-desired product features

Publications, presentations, and reports: -- Two (2) conference papers -- One (1) journal paper -- One (1) thesis -- One (1) proposal Datasets and algorithms: -- One (1) algorithm (code with associated training and testing data)

Objective 3.5.b: Network-based modeling of customer preference incorporating marketing parameters

Objective 3.5.c: Design of marketing strategies with fairness consideration and validate the approach

Goal 3.6 (SA6) Staff: Huang, Li, Mary, Ussery, CI ARP team

Objective 3.6.a: Design and develop machine learning algorithms and software, and advanced security and privacy technologies, for privacy-preserving data analytics.



Research Theme


Objective 3.6.b: Train, test and validate the models and algorithms with publicly available data and some controlled genomics and health data; develop innovative frameworks and practical privacy-preserving techniques.

Datasets and algorithms: -- One (1) algorithm (code with associated training and testing data)

Objective 3.6.c: Test the algorithms and technologies to work with a wide range of data types and high-dimensional heterogeneous data sources; Develop and deploy bioinformatics workflows into the private cloud environment, the Arkansas Research Platform ARP.

Goal 3.7 (SA7)Staff: Qinghua Li; Xiuzhen Huang; Ningning Wu

Objective 3.7.a: Develop privacy-preserving federated learning methods through combining cryptography techniques and privacy models

Publications, presentations, and reports:-- Two (2) conference papers-- One (1) journal paper-- One (1) thesis-- One (1) proposal


Research Theme


Objective 3.7.b: Explore how to protect the privacy of classification input data from the server hosting machine learning models

Datasets and algorithms:-- One (1) algorithm (code with associated training and testing data)

Objective 3.7.c: Assess/Protect the trustworthiness of training data and machine learning models

Research Theme 4: Social Media and Networks

Goal 4.1 (SM1) Staff: Zhan, Adams, S. Yang

Objective 4.1.a: Develop a cyber discourse social network platform

Publications, presentations, and reports: -- Five (5) peer-reviewed journal and/or conference papers (articles) Assessments, questionnaires, and surveys: -- One (1) IRB-approved questionnaire for collecting discourse data Applications and platforms: -- One (1) cyber discourse social network platform

-- Cyber-discourse platform sees use within and outside of DART -- DART Git repository accesses, downloads, and branches show acceptance and use of algorithms within DART -- Public Git repository accesses, downloads, and branches show adoption of algorithms in the data science community

-- Cyber discourse platform is used for social media and network discourse data collection and analysis --Policies and public debate consider deviant cyber campaigns --New studies in this research discipline cite datasets provided by this research -- Perceptual information is considered in the development of new smart technologies designed to influence social and ethical

Objective 4.1.b: Collect data using the developed cyber discourse social network platform


Research Theme


Objective 4.1.c: Develop natural language processing algorithms to analyze discourse data collected by the platform as well as existing data

Datasets and algorithms: -- Two (2) advanced natural language algorithms (code with associated training and testing data)

behavior --Realtime, disaster-relevant data posted to social platforms and collected from various imagery sources see incorporation into disaster response planning.

Goal 4.2 (SM2) Staff: Agarwal, Trudeau, Zhan, Milburn, Dagtas Equipment: COSMOS Data Servers Infrastructure: Social media data acquisition license

Objective 4.2.a: Characterize online information environment (OIE)

Publications, presentations, and reports: -- One (1) taxonomy -- One (1) journal paper -- Two (2) conference presentations Applications and Platforms: -- Socio-computational models for OIE and TTP implemented in a web-based application Datasets and algorithms: -- Two (2) socio-economic models and associated datasets

-- Citations of papers, presentations, and reports -- Socio-computational platform sees use within and outside of DART -- DART Git repository accesses, downloads, and branches show acceptance and use of algorithms within DART -- Public Git repository accesses, downloads, and branches show adoption of algorithms in the data science community -- Findings are incorporated in courses taught by the PI imparting new skills to analyze social media and social networks. -- Datasets are accessed by academics and industry under NSF guidelines and terms of

Objective 4.2.b: Develop socio-computational models to identify key actors and key groups of actors

Objective 4.2.c: Study tactics, techniques, and procedures (TTPs) of deviant cyber campaigns

Objective 4.2.d: Develop socio-computational models to measure power of a cyber campaign


Research Theme


service agreement with social media companies

Goal 4.3 (SM3)Staff: Dagtas, Trudeau, MilburnPartnerships: UALR, UA Law School

Objective 4.3.a: Develop multimedia indexing methods for social media data

Publications, Presentations, and Reports-- Two (2) journal articles-- One (1) conference paperDatasets and Algorithms:-- Three (3) algorithms ---- 1 x Indexing---- 1 x deep learning for multimedia data---- 1 x integrated smart applications

-- Citations of papers, presentations, and reports-- Socio-computational platform sees use within and outside of DART-- DART Git repository accesses, downloads, and branches show acceptance and use of algorithms within DART-- Public Git repository accesses, downloads, and branches show adoption of algorithms in the data science community-- Findings are incorporated in courses taught by the PI imparting new skills to analyze multimedia data and other unstructured data

Objective 4.3.b: Design and implement deep learning methods for multimedia data

Objective 4.3.c: Build Integrated smart applications based on unstructured multimedia data

Goal 4.4 (SM4) Staff: Milburn, Dagtas, Liao, Zhan, Cothren, Ussery, Talburt,

Objective 4.4.a: Extract and index content describing transportation infrastructure status from social platforms

Publications, presentations, reports: -- Three (3) journal articles -- Two (2) conference papers

-- Citations of papers, presentations, and reports -- Socio-computational platform sees use within and outside of DART


Research Theme


Nachtmann, Rainwater, Celebi, Karim

Objective 4.4.b: Fuse data from social platforms describing transportation infrastructure status with other data sources

-- Two (2) case studies of natural disasters scenerios Applications and Platforms: -- GIS routing platform infromed with social media feeds Datasets and Alogirthms: -- One (1) schema for mapping each datum to a probability describing its credibility -- Four (4) algorithms

-- DART Git repository accesses, downloads, and branches show acceptance and use of algorithms within DART -- Public Git repository accesses, downloads, and branches show adoption of algorithms in the data science community -- GIS platform is used by other researchers to better understand how social media input can be used dynamically for better routing

Objective 4.4.c: Assess credibility of data inputs from Objectives 4.4.a and 4.4.b

Objective 4.4.d: Develop routing algorithms that use inputs from Objectives 4.4.a-4.4.c to support routing for disaster response

Research Theme 5: Learning and Prediction

Goal 5.1 (LP1) Staff: Liu, Chimka

Objective 5.1.a: Create the Random Forests for Recurrent Event Analytics, which integrates the RF algorithm with classical statistical methods allows dynamic feature information to be

Publications, presentations, reports: -- Two (2) manuscripts submitted for publication -- Two (2) student theses or dissertations proposed -- Two (2) student theses or dissertations defended

New statistical and machine learning paradigms; Formal comparison of ensemble learning with traditional methods.

-- Documented understanding of complex/dynamic relationship between event processes and covariates information; Proposed interpretable intervention, control and optimization actions


Research Theme


incorporated into a tree-based method.

-- One (1) submitted research proposals

related to recurrent event analysis--Promote acceptance of AI algorithms in on-the-edge devices and low-cost computers --Teach new lecture content in Deep Learning; -- Demonstrate reduced computational complexity and improve performance accuracy in AI -- Create research environment in higher education that trains students to experience and execute real-world AI applications -- Intervention, control and optimization policies via improved feature utilization; New statistical learning-based research practices; Research findings shared with industry to support practice of regular and rigorous data collection -- Motivate use of the integrated framework to improve decision making in business problems

Objective 5.1.b: Create the Gradient Boosting method for Recurrent Event Analytics, which integrates the boost trees with classical statistical methods allows dynamic feature information.

Objective 5.1.c: Perform comparison study between the methodologies above and identify future research directions

Goal 5.2 (LP2) Staff: Rainwater, Liu

Objective 5.2.a: Develop methodology integrating the marked temporal point process (MTPP) with long short-term memory networks (LSTM)

Publications, presentations, reports: -- One (1) conference paper -- Two (2) conference presentations -- Two (2) journal publications -- One (1) case study Workshops, demonstrations, and trainings:

Documented improved integration of large-scale computing knowledge with marked temporal point process (MTPP) and deep learning; Curation of discrete marked temporal point process (MTPP) data set; Benchmark results comparing marked temporal point process

Objective 5.2.b: Create scalable implementation of MTTP/LSTM approach applicable to real-world data analysis scenario


Research Theme


Objective 5.2.c: Evaluate and assess MTTP/LSTM approach on real-world discrete data sets

-- One (1) graduate seminar -- One (1) industry workshop

(MTPP) against other deep learning approaches

Goal 5.3 (LP3) Staff: Celebi, Kursun, Luu, Kim, Karim

Objective 5.3.a: Extract explanatory features from Deep Network

Publications, presentations, reports: -- Six (6) conference papers -- Two (2) Journal publication -- Two (2) Master's theses; Workshops, demonstrations, and trainings: -- One (1) workshop -- One (1) teaching module -- One (1) Special topics undergraduate class offering Datasets and Alogrithms: -- One (1) dataset

Identification of explanatory features in Deep Networks; Demonstrated ability to design improved reward function and explain causal relationships in DRL; Training of graduate and undergraduate students

Objective 5.3.b: Address high dimensionality issues in Deep Reinforcement Learning (DRL) using algebraic and topological methods

Objective 5.3.c: Designing a novel rewarding model, and addressing interpretability issues in DRL

Goal 5.4 (LP4) Staff: Khoa Luu, Ngan Le

Objective 5.4.a: Create Novel Deep Learning Networks Executable with Reduced Computational Resources and Assess Performance

Publications, presentations, reports: -- One (1) journal article -- Two (2) conference papers -- One (1) invited

Train graduate students and honors undergraduate students; Disseminate new low-computational cost Deep Learning algorithms,


Research Theme


Objective 5.4.b: Address Low-cost Deep Learning Algorithmic Analysis and Challenges

presentation at an Arkansas Institution -- One (1) invited presentation elsewhere Workshops, demonstrations, and trainings: -- One (1) conference workshop -- One (1) tutorial -- One (1) teaching module

frameworks, and supportive libraries; Contribute new dimensions to analyze the computational time, resource consumption, and performance in AI algorithms

Objective 5.4.c: Explore Low-cost Deep Learning Applications in Natural Images and Medical Images

Goal 5.5 (LP5) Staff: Zhang, Nachtmann

Objective 5.5.a: Design advanced feature engineering techniques for high-dimensional temporal data

Publications, presentations, reports: -- Three (3) journal publications -- Three (3) conference presentations -- One (1) doctoral dissertation

Improve knowledge in feature engineering with transaction data; Establish abilities to incorporate feature engineering in prediction; Employ the skills of the integrated framework to solve business problems

Objective 5.5.b: Create an improved prediction and decision-making framework incorporating feature engineering with health transaction data

Objective 5.5.c: Employ and validate the new framework for prediction and decision making with business transaction data


Research Theme


Research Theme 6: Education

Goal 6 (ED)Staff: Fowler, Schubert, Addison

Objective 6.1.a: Middle school coding block

9 week curriculum developed in consideration of standards, piloted in at least 6 schools

Students successfully complete coding block and increase in enrollment in computer science courses at high school level.

Objective 6.1.b: Create model postsecondary programs in 3 phases at collaborating institutions

Curriculum implemented at three cohorts of 3+ campuses each,

Curriculum has been implemented on two- and four-year campuses. Programs are beginning to produce graduates who have completed the new curriculum.

Research Theme 7: Workforce Development

Goal 7 (WD) Staff: Fowler, Schubert, Addison

Objective 7.1.a: K12 Teacher PD

4 training workshops and 8 webinars, platform for resource sharing and troubleshooting

400+ K12 teacher participants confidently teach tech and leadership skills, 4+ students present at EAST conference

Objective 7.1.b: Education & Broadening Participation Seed Grants

TBD TBD TBD

Objective 7.1.c: Provide funding for faculty at collaborating institutions to learn new skills and tools in data science

50 faculty trained Developing human infrastructure needed to achieve implementation

Each collaborating institution has faculty resources needed to teach DS curriculum

Objective 7.1.d: Career Development Workshops

15 workshops hosted Increased research competitiveness with progress in proposals submitted and collaborative projects


Research Theme


Objective 7.2.a: Student Support at Participating Institutions

75+ undergraduates supported and trained, 75+ graduate students supported and trained

Network of student participants established

Students with rich educational experiences matriculating and getting jobs in relevant industries

Objective 7.2.b: Summer Internships

20+ internships placed Students are aware of the opportunities and there is completion for placements.

Interns graduating and entering workforce

Objective 7.2.c: Research-based capstone projects

9+ Capstone project documentation disseminated to ARHE community

Research faculty are contributing a variety of potential capstone projects. Students have capstone opportunities relevant to their chosen specialization in data science.

Students are graduating with meaningful and relevant capstone experiences.

Objective 7.2.d: Arkansas Summer Research Institute

5 ASRI workshops hosted, 200+ students trained

Positive evaluations and feedback from student participants

Students with enhanced DS/CS skills matriculating through colleges and universities

rii track-1: data analytics that are robust and trusted

Documents